Exploring the Synergy Between Ethereum Layer 2 Solutions and Machine Learning to Improve Blockchain Scalability

Abstract

Blockchain technologies, despite their profound transformative potential across multiple industries, continue to face significant scalability challenges. These limitations are primarily observed in restricted transaction throughput and elevated latency, which hinder the ability of blockchain networks to support widespread adoption and high-volume applications. To address these issues, research has predominantly focused on Layer 1 solutions that seek to improve blockchain performance through fundamental modifications to the core protocol and architectural design. Alternatively, Layer 2 solutions enable off-chain transaction processing, increasing throughput and reducing costs while maintaining the security of the base layer. Despite their advantages, Layer 2 approaches are less explored in the literature. To address this gap, this review conducts an in-depth analysis on Ethereum Layer 2 frameworks, emphasizing their integration with machine-learning techniques, with the goal of promoting the prevailing best practices and emerging applications; this review also identifies key technical and operational challenges hindering widespread adoption.

1. Introduction

Since its introduction in 2009, Bitcoin has sparked a major shift in the digital world by pioneering blockchain technology to securely record transactions without relying on a central authority [1]. As the first successful cryptocurrency, Bitcoin uses blockchain to keep an immutable record of financial transfers while preventing issues like double-spending [2]. Instead of a single trusted party controlling the ledger, blockchain distributes this role across all participants, who work together through a decentralized consensus process to validate and order transactions. This collective agreement ensures transparency and security, enabling trust in the system without intermediaries. The impact of this breakthrough extends far beyond Bitcoin, reshaping how data is managed and verified across many industries [3,4,5]. Unlike Bitcoin [5], which primarily focuses on peer-to-peer digital currency, Ethereum took a different approach by pioneering the concept of smart contracts, which are automated programs that execute specific actions when triggered by transactions on the blockchain [6]. This innovation allows for much more complex and flexible interactions, enabling users to create a variety of decentralized applications (dApps) that go far beyond simple payments. By embedding code directly into the blockchain, Ethereum has expanded the possibilities of what can be achieved with decentralized technology [7], opening up new opportunities across many fields. This growth has led to scalability issues for prominent public blockchain platforms [8], exerting considerable pressure on network resources and resulting in increased transaction latency, thereby impeding widespread adoption. To facilitate the widespread adoption of blockchain, more advanced and sophisticated scalability solutions will be necessary in the future [9]. These solutions could involve machine learning [10] and artificial intelligence (AI) [11,12,13].

Our focus is on reviewing existing Layer 2 scalability solutions for the Ethereum blockchain. Figure 1 illustrates a hierarchical model of blockchain architecture, starting from Layer −1 and progressing to Layer 2 [14]. At the foundation, Layer −1 consists of the physical hardware infrastructure that powers the blockchain network. Above this, Layer 0 comprises the peer-to-peer network of nodes responsible for data exchange and communication across the system. Layer 1 represents the core blockchain protocol, where consensus mechanisms, data structures, and on-chain transactions are executed. Building on top of this, Layer 2 introduces scalability solutions that operate independently of the main chain. These protocols enable off-chain transaction processing through secure and authenticated communication channels, reducing the load on Layer 1 while improving throughput, efficiency, and cost-effectiveness.

1.1. Related Work

Among the various obstacles to broader implementation, scalability emerges as a fundamental challenge hindering the deployment of public blockchains in real-world business environments. Numerous researchers have explored this issue and proposed a variety of solutions. Several studies have systematically examined these scalability challenges and proposed potential solutions, as summarized below.

In [9], Zhou et al. systematically categorize and analyze existing scaling solutions at different levels, comparing their effectiveness in addressing issues such as low throughput and high transaction latency. Expanding on this analysis, Sanka and Cheung [15] provide a comprehensive literature review, highlighting the inherent trade-offs between scalability, security, and decentralization.

In [15], Sanka and Cheung conduct an extensive literature review on blockchain scalability, highlighting key limiting factors, such as low throughput; high latency; and the inherent trade-offs between scalability, security, and decentralization. Their study also assessed existing approaches aimed at addressing these challenges. On-chain methods like increasing block size or data reduction offer limited throughput improvements and pose risks such as centralization and security vulnerabilities. Overall, these limitations underscore that no single solution fully addresses all scalability challenges without compromising other key blockchain properties. Beyond general scaling techniques, state channels have emerged as a promising method to address persistent latency and fee issues, as explored by Negka and Spathoulas [16].

In [16], the authors present an in-depth evaluation of state channels as a promising approach to enhance blockchain scalability by addressing persistent issues, such as network latency and excessive transaction fees. Their study systematically reviews existing research, outlining a variety of state channel architectures and their operational mechanisms aimed at increasing transaction throughput while preserving security. Despite their potential, state channels face significant challenges, including limited interoperability across different blockchain platforms, security risks inherent to off-chain transactions, and the complexity involved in their deployment and maintenance. The authors provide a comparative analysis of these challenges across different implementations, highlighting critical trade-offs and technical hurdles. They recommend that future investigations focus on advancing interoperability solutions, strengthening security measures, and enabling seamless integration of state channels into broader blockchain infrastructures. This comprehensive review contributes essential knowledge toward the maturation and practical adoption of state channel technologies in real-world blockchain systems. While state channels offer certain advantages, rollups have been proposed as an alternative solution to further enhance throughput and efficiency [17].

Thibault et al. presents, in [17], a comprehensive analysis of rollups as a critical mechanism for addressing scalability limitations in blockchain systems. Their study examines the architectural frameworks, practical deployments, and operational trade-offs associated with rollup technologies. Although rollups substantially alleviate issues related to low transaction throughput, the review highlights persistent limitations, including residual latency, security concerns, and integration challenges. Comparative assessments reveal considerable variation in the performance of different rollup models, suggesting a need for further refinement. The authors emphasize several underdeveloped aspects, particularly in terms of interoperability and user experience. To advance the robustness and adoption of rollup-based solutions, they support the development of more secure architectures, improved cross-layer compatibility, and enhanced execution efficiency. Building on these rollup mechanisms, Yi [18] provides a broader overview of Layer 2 blockchain solutions and their role in enabling fast, low-cost transactions.

The review conducted by Yi in [18] examines Layer 2 blockchain technologies as vital enhancers of scalability and efficiency for dApps, emphasizing their operation atop base blockchains to facilitate fast, low-cost transactions while maintaining security and decentralization. It offers a comprehensive analysis of foundational concepts and platform-specific implementations, primarily on Bitcoin and Ethereum, highlighting both advantages and inherent limitations. Key constraints include interoperability challenges and the complexity of seamless integration with base layers. Future research directions propose advancements in cross-chain interoperability, enhanced privacy-preserving methods, and optimization of smart contract execution to further transform decentralized ecosystems and improve user engagement. Complementing these Layer 2 strategies, recent research has explored the integration of AI and machine learning to further optimize scalability, security, and privacy in blockchain systems [12].

In [12], Yuan et al. investigates the application of AI techniques, including machine learning and deep learning, to enhance blockchain’s scalability, security, and privacy. It emphasizes AI-driven optimization of consensus protocols, smart contract vulnerability detection, and privacy-preserving methods such as federated learning (FL) and differential privacy. Limitations include the lack of deep theoretical integration between AI and blockchain and challenges in applying AI methods universally across diverse blockchain environments. Future research has proposed the development of more unified frameworks for AI–blockchain synergy, expanding privacy techniques, and advancing adaptive security mechanisms to address evolving threats and scalability demands. In addition to general blockchain optimization, AI and machine learning have been applied to domain-specific problems such as fraud detection in FinTech systems [19].

Hanae et al. [19] examines the integration of machine-learning and blockchain technologies to enhance transactional fraud detection within FinTech systems, highlighting a novel framework that synergizes predictive analytics with blockchain’s secure ledger to combat sophisticated and evolving fraudulent activities. Limitations include the potential complexity of system integration; scalability concerns; and the need for extensive, high-quality datasets to train predictive models effectively. Future research should address the development of adaptive, real-time fraud-detection mechanisms; improve interoperability between diverse blockchain platforms; and explore transparent, explainable AI approaches to bolster trust and efficacy in fraud mitigation.

Table 1. Summary of the related work.

Reference	Authors	Focus Area	Key Contributions	Limitations	Future Directions
[9]	Zhou et al.	Scalability survey	Analyzes scalability methods for throughput and latency.	No single method resolves scalability without trade-offs.	Develop integrated multi-technique solutions.
[15]	Sanka and Cheung	On-chain scaling	Highlights trade-offs in scalability, security, decentralization.	On-chain methods risk centralization and insecurity.	Explore hybrid models balancing scalability and security.
[16]	Negka and Spathoulas	State channels	Reviews state channels reducing latency and costs securely.	Faces interoperability and security challenges.	Improve interoperability, security, and integration.
[17]	Thibault et al.	Rollups	Studies rollups’ role in scaling blockchain.	Suffers from latency, integration, and interoperability issues.	Enhance security, compatibility, and efficiency.
[18]	Yi	Layer 2 dApps	Examines Layer 2 scalability on Ethereum and Bitcoin.	Integration and cross-chain interoperability remain hard.	Advance interoperability, privacy, and contract optimization.
[12]	Yuan et al.	AI–blockchain	Explores AI for optimizing consensus, security, and privacy.	Limited AI–blockchain integration and scope.	Build unified AI–blockchain frameworks and adaptive security.
[19]	Hanae et al.	Machine learning and blockchain for fraud detection	Proposes fraud detection using AI and blockchain.	Complex systems, scalability limits, data dependency.	Create adaptive real-time detection with explainable AI.

provides a concise comparison of selected related works, summarizing their focus areas, key contributions, identified limitations, and proposed directions for future research in the context of blockchain scalability and optimization.

1.2. Contributions

The contribution of this review can be summarized as follows:

• Conducting a comprehensive analysis of recent and State-of-the-Art research on Ethereum Layer 2 scalability solutions and their integration with machine-learning techniques.
• Categorizing Layer 2 protocols based on their design principles.
• Exploring practical use cases where Layer 2 technologies and machine learning intersect to enhance blockchain performance.

1.3. Objectives

This review aims to systematically explore and analyze current research on the integration of Ethereum Layer 2 scalability solutions with machine learning. It examines a wide range of articles, providing a comprehensive overview of the field’s development. The review highlights the goals, assumptions, and benefits of various Layer 2 protocols while evaluating their compatibility with machine learning. It also identifies key challenges hindering their integration.

1.4. Research Questions

In accordance with the objectives of this systematic review, the following research questions were formulated to direct the focus of the study:

• RQ1: How have scalability issues been addressed in the Ethereum blockchain?
• RQ2: What are the key types of Ethereum Layer 2 scaling solutions currently implemented, and how do they differ in terms of architecture, security, and scalability?
• RQ3: How can machine-learning techniques be integrated with Layer 2 solutions to further optimize blockchain scalability and performance?

1.5. Paper Outline

The remainder of this paper is structured as follows. Section 2 provides an overview of blockchain trilemma. Section 3 outlines the research methodology adopted for this review. Section 4 presents the data analysis, while Section 5 discusses how the findings address the research questions. Section 6 discusses the results and Section 7 concludes the paper and proposes future research directions.

2. From Blockchain Trilemma to Blockchain Quadrilemma

The blockchain scalability trilemma, first described by Ethereum co-founder Vitalik Buterin [20], suggests that a blockchain network can effectively optimize only two out of three key properties: scalability, security, and decentralization [8]. Enhancing all three simultaneously poses a significant technical challenge. For instance, increasing scalability often requires more centralized control or compromises on security, while maximizing decentralization and security can lead to slower transaction speeds and limited throughput. This trilemma has driven ongoing innovation in blockchain architecture, with various platforms exploring approaches like sharding [21,22], Layer 2 solutions [23,24], sidechains [25,26] and novel consensus algorithms [27,28] in an effort to balance these three goals.

Scalability remains a challenge for traditional blockchains, such as Bitcoin and Ethereum, due to the growing number of users and transactions. Algorand presents itself as a blockchain solution that addresses the long-standing challenge of balancing scalability, security, and decentralization [29].

The widely accepted concept of the blockchain scalability trilemma expanded into a quadrilemma, as illustrated in Figure 2. Trust has been introduced as a fourth factor [15], extending the trilemma into a quadrilemma, due to its significant influence on scalability.

Trust, while essential for scaling blockchain systems, often exists in tension with decentralization. Systems that rely on trusted entities can simplify consensus protocols, communication, and computation, thereby achieving greater scalability. However, this simplification may come at the cost of decentralization and, potentially, security. This trade-off has led to the formulation of the blockchain scalability quadrilemma, which underscores the difficulty of optimizing scalability, security, decentralization, and trust simultaneously in current blockchain architectures [30]. For instance, private and consortium blockchains can attain high levels of security and scalability due to their reliance on trusted parties, though they tend to be centralized. Public blockchains, while strong in both decentralization and security, typically suffer from limited scalability. Hence, designing an optimal blockchain solution requires balancing these four attributes to achieve a practical trade-off [31].

Blockchain scalability is typically assessed using key performance indicators, such as transaction throughput and latency for write operations, data read throughput and latency for read operations, and storage capacity for evaluating storage performance. Other important metrics include the transaction success rate and the level of resource consumption, which takes into account factors like central processing unit (CPU) usage and network traffic [15].

3. Methodology

This systematic literature review was conducted based on the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) [32], ensuring methodological rigor and alignment with the computer science domain. Figure 3 illustrates the complete methodology adopted in this review. The PRISMA approach involved the following systematic steps:

• Identification, which entailed developing a comprehensive search strategy to locate all relevant articles by querying multiple academic databases using carefully selected keywords and Boolean operators tailored to the research topic.
• Screening, where all retrieved articles were assessed to remove duplicates and exclude those that did not meet the predefined relevance or exclusion criteria.
• Inclusion, whereby, after a thorough eligibility assessment, only those articles that satisfied all inclusion criteria were confirmed for incorporation into the systematic review.

Figure 3. PRISMA flow—the article-selection process.

Figure 3. PRISMA flow—the article-selection process.

In addition, we plot the number of papers published per year to reveal temporal trends, and we illustrate this in Figure 4.

The PRISMA diagram depicts the article selection process, as detailed below:

• Identification: Articles were retrieved from five electronic databases, namely MDPI (4 articles), IEEE Xplore (39 articles), ScienceDirect (6 articles), Scopus (17 papers), and Web of Science (26 papers), as well as other sources (111 articles).
• Screening: In the initial phase, the titles and abstracts of the retrieved articles were screened to assess their relevance to blockchain applications in diploma management, resulting in the exclusion of 58 articles that addressed unrelated topics. Furthermore, 102 articles were excluded based on the predefined exclusion criteria.
• Inclusion: After applying the screening criteria, a final set of 20 studies was selected for detailed analysis.

A comprehensive description of the article-selection methodology can be found in Section 3.1, Section 3.2, Section 3.3, Section 3.4, Section 3.5 and Section 3.6.

3.1. Selection of Primary Studies

The search was conducted across titles, abstracts, and keywords. To identify relevant primary studies, the following search keywords were used: “blockchain” AND “scalability” AND “machine learning” AND (“layer 2” OR “layer-2” OR “off-chain” OR “Layer-two”). For each retrieved paper, the title was first evaluated, followed by a review of the abstract, introduction, and conclusion to determine compliance with the predefined inclusion criteria. Subsequently, a thorough analysis of the full text was conducted to identify key contributions and open research issues.

3.2. Selection Results

Considering the keywords, the initial results of this search are shown in

Table 2. Initial search results.

Database	Where the Search String Was Applied	Results
MDPI	Title/keyword	4
IEEE Xplore	Publications	39
ScienceDirect	Title, abstract, keywords	6
Web of Science	Title, abstract, keywords	26
Scopus	Publications	17
Other sources	Publications	111

.

3.3. Inclusion and Exclusion Criteria

The inclusion and exclusion criteria in this systematic literature review are designed to specifically accept only those documents or articles that directly address the current state of Ethereum Layer 2 solutions, possibly integrating some machine-learning techniques. The criteria include the following:

• The paper must address blockchain scalability Layer 2 solutions combined with machine learning, either directly or indirectly, and identify relevant causes or contributing factors.
• The paper must propose a feasible solution (e.g., method, technique, model, or framework) aimed at addressing blockchain scalability, Layer 2 scaling solutions, and machine learning.
• The paper must be published in peer-reviewed journals or conference proceedings.
• The paper must be written in English.

3.4. Data Extraction

The data sources selected encompass both broad and specialized computer science and multidisciplinary databases to ensure comprehensive literature coverage. Each article was individually assessed for scholarly integrity and peer-review status. Articles were retrieved from the following sources:

• MDPI—A well-established platform that facilitates scientific exchange and offers an extensive repository of articles with advanced keyword- and topic-based search functionalities.
• IEEE Xplore—A comprehensive digital library providing access to a wide range of technical literature in engineering, computer science, and related fields.
• Scopus—A leading abstract and citation database covering diverse disciplines.
• Web of Science—A multidisciplinary citation index recognized for high-quality peer-reviewed sources.
• ScienceDirect—A platform offering access to a broad collection of scientific and technical research articles.
• Other sources—Obtained through ResearchGate, where original sources were identified and verified, as the platform primarily serves as a researcher networking site and often contains duplicates of peer-reviewed publications. We identified and included articles from peer-reviewed journals such as the Journal of Big Data, Cybernetics and Computer Technologies, IET Blockchain, International Journal of Innovation Management and Organizational Behavior, International Journal for Research in Applied Science and Engineering Technology, and the Communication and Information Technology Journal, each providing valuable contributions to academic research, open-access publishing, and interdisciplinary studies across diverse fields.

3.5. Eligibility Criteria

This review systematically collects and analyses recent research concerning scalability issues of blockchain technology and the applicability of Layer 2 solutions for this issue. Furthermore, machine-learning techniques could be possibly integrated. To ensure relevance to current developments, only studies published between 2020 and 2025 were considered.

In order to uphold the quality and credibility of the review, only articles published in peer-reviewed journals or reputable conference proceedings were included. This selection criterion guarantees that the sources have undergone rigorous academic scrutiny, thereby enhancing the validity and reliability of the review’s findings. Articles published in languages other than English were excluded to facilitate a consistent and comprehensive analysis.

Moreover, included studies were required to have full-text availability online. This criterion ensures that the review is based on detailed content, allowing for an in-depth assessment of each article’s methodology, results, and conclusions. Limiting inclusion to studies with accessible complete texts mitigates potential biases associated with reliance on partial information, such as abstracts or titles alone.

The specific inclusion and exclusion criteria guiding the article selection are summarized in

Table 3. Inclusion and exclusion criteria.

Inclusion Criteria	Exclusion Criteria
Analyses blockchain technology and Layer 2 scaling solutions or possibility of synergy with machine learning	Publications released outside the 2020 to 2025 timeframe
	The articles are not published in journals and conference papers
	Articles written in languages other than English
	Articles for which the full text was not accessible

.

The number of included and excluded papers can be found in Figure 5.

3.6. Screening and Selection

To ensure the relevance of each article to the research questions, an incremental screening approach was employed. The first step involved removing duplicate entries retrieved from multiple data sources. Subsequently, the titles of all remaining papers were meticulously reviewed to exclude those not aligned with the scope of this review, specifically papers that addressed blockchain in general but did not focus on Layer 2 solutions or machine learning. In cases where title-based screening was inconclusive, the abstract was examined in detail to determine the article’s suitability. The predefined inclusion and exclusion criteria played a crucial role in systematically assessing each paper for its relevance to the research questions.

The distribution of the included articles by year of publication is depicted in Figure 6. All studies incorporated in this review were published within the last five years, underscoring this systematic review’s focus on recent advancements.

Following the inclusion and exclusion criteria, Figure 7 depicts the number of excluded papers.

To ensure that the analysis accurately represents the most recent advancements and perspectives in the field, only articles published between 2020 and 2025 were considered for inclusion. This focus on the contemporary literature allowed for the prioritization of the latest findings and developments. Consequently, 6 articles published outside this specified timeframe were excluded from the review.

A number of 58 papers that addressed blockchain in general but did not focus on Layer 2 solutions or machine learning were excluded.

Additionally, 25 articles were omitted due to not meeting the requirement of being published in peer-reviewed journals or presented at reputable conferences, thus maintaining the academic rigor of the systematic review.

Among the screened studies, 2 articles written in other languages were excluded, while all remaining articles were published in English.

Furthermore, inclusion criteria mandated that the full text of each article be accessible online to enable a comprehensive evaluation of the study. Accordingly, 67 articles were excluded due to the unavailability of their complete content online.

The distribution of the included articles across the consulted databases is illustrated in Figure 8. In total, three articles were retrieved from MDPI, IEEE Xplore, ScienceDirect, and Scopus, each, while Web of Science contributed two articles. An additional six articles were obtained from other peer-reviewed journals, including the Journal of Big Data, Cybernetics and Computer Technologies, IET Blockchain, International Journal of Innovation Management and Organizational Behavior, International Journal for Research in Applied Science and Engineering Technology, and the Communication and Information Technology Journal.

In order to strengthen the comprehensiveness and validity of the review, we examined the reference lists of the selected studies, with particular emphasis on the most frequently cited publications within the field. This approach ensured that seminal and influential contributions were not overlooked and that the review incorporated a balanced representation of the existing body of knowledge. By systematically including these highly cited works, the potential risk of selection bias was further minimized, thereby enhancing the reliability and robustness of the findings.

A total of 20 studies met the inclusion criteria and were included in the systematic literature review for the qualitative synthesis phase. Collectively, these publications provide significant academic contributions to understanding the synergy between Layer 2 scaling solutions and machine learning.

4. Results

This article presents a systematic examination of scholarly works that explore the application of blockchain technology and Layer 2 scaling solutions, alongside the potential integration with machine learning. Focusing on studies that investigate these technologies in diverse implementation phases, this review incorporates 20 selected publications, following a rigorous screening process. These sources provide critical perspectives on the advancements, challenges, and synergies emerging at the intersection of blockchain scalability and intelligent data processing.

To summarize, referring to Figure 3, a total of 203 records were retrieved from the selected databases. During the initial screening, 25 duplicate records were removed, followed by the exclusion of 58 studies that were unrelated to Layer 2 scalability solutions or blockchain and machine-learning synergy. Subsequently, five additional articles were identified through reference-list examination of highly cited papers and Google Scholar searches. Among these, three articles were excluded due to their scope. After applying the eligibility criteria, 102 articles were further excluded. Ultimately, 20 studies satisfied the defined eligibility requirements and were incorporated into this systematic review.

Figure 9 presents an overview of all the sources retrieved for this study prior to the screening and selection process. The collected records encompass a diverse range of publication types, including journal articles, conference papers, preprints, book chapters, research reports, patents, and theses, reflecting the broad scope of the literature considered in this review.

The distribution of selected articles by publication type is presented in Figure 10. As shown, the majority of the included studies (85.0%) were published in peer-reviewed scholarly journals, highlighting the central role of the high-quality academic literature in this field. The remaining 15.0% comprised published conference papers. This distribution reflects the predominance of rigorously reviewed research disseminated through reputable journals within the scope of this systematic review.

Table 4. Selected relevant papers.

Reference	Research Title	Authors	Summary of the Research
[33]	“Trustworthy Reputation for Federated Learning in O-RAN Using Blockchain and Smart Contracts”	Javed et al.	The paper proposes a blockchain-based framework for trustworthy federated learning in O-RAN, tackling multi-vendor trust and privacy issues with an on-chain reputation system. Smart contracts automate participant registration, model verification, and reputation scoring for tamper-proof accountability. Implemented on Polygon Layer 2 with a blockchain oracle, the design integrates with O-RAN via a dApp and aims to leverage future scalability improvements.
[14]	“A survey of Layer-two blockchain protocols”	Gangwal et al.	The paper reviews Layer 2 blockchain protocols as scalable solutions that address the limitations of Layer 1 blockchains like low transaction throughput and high latency. Layer 2 protocols improve performance by executing most transactions off-chain while relying on the main chain for security and dispute resolution. The study offers a detailed taxonomy; compares different Layer 2 types; and discusses their mechanisms, challenges, and potential, providing a comprehensive overview of current advancements in blockchain scalability.
[34]	“A Survey on Data Availability in Layer 2 Blockchain Rollups: Open Challenges and Future Improvements”	Saif et al.	The paper surveys Layer 2 blockchain rollups as scalable solutions enhancing transaction speed, security, and efficiency. It highlights the data-availability challenge, stressing the importance of reliably posting off-chain data on-chain to prevent attacks and enable verification. The study reviews current rollup designs, their strengths and weaknesses, and suggests future research to improve Layer 2 scalability and reliability.
[35]	“Architecture for the Academic Certificate System on the Ethereum Layer 2 Solution”	Wanotayapitak	The research addresses Ethereum’s scalability issues like network congestion and high fees by proposing a Layer 2 architecture for academic certificate systems. It identifies key technologies such as Interplanetary File System (IPFS), Oracle, and Decentralized Identifiers, with IPFS being most popular. Among Ethereum Layer 2 solutions, Arbitrum ranks highest in performance, followed by Polygon and Optimism. The proposed architecture is implemented to demonstrate practical viability and provides a flexible blueprint for building efficient dApps on Ethereum Layer 2 networks, with applicability beyond academic certification. This work contributes valuable insights into scalable blockchain deployment for dApps.
[36]	“Assessing the Environmental Sustainability of Polygons Consensus Mechanism and Transaction Processing, Comparing Its Energy Consumption and Carbon Footprint with Other Layer 2 and Layer 1 Blockchain Solutions, And Exploring Potential Avenues for Further Optimization”	Paul et al.	The research evaluates Polygon’s energy use and carbon footprint, showing that its proof-of-stake consensus greatly reduces emissions compared to proof-of-work consensus. Following Ethereum’s transition to proof-of-stake consensus, Polygon’s carbon footprint dropped by over 99%, making it one of the most sustainable blockchains. The study also explores renewable energy integration and governance for further optimization.
[37]	“Designing an Intelligent System for Vaccine Supply Chain Management Based on Blockchain Using Machine Learning Algorithms”	Ahmadi et al.	The study proposes an intelligent vaccine supply chain management system integrating blockchain technology, machine learning, and Internet-of-Things devices. It uses a hybrid Long Short-Term Memory model for manufacturer credibility assessment and a Support Vector Machine module for vaccination prediction, with blockchain sharding to enhance scalability. Tested on real and simulated data, the system securely tracks vaccines, detects expired doses, and manages high transaction volumes, improving supply chain reliability, safety, and scalability for global immunization efforts.
[38]	“Bio-Rollup: A new privacy protection solution for biometrics based on two-layer scalability-focused blockchain”	Yun et al.	The paper introduces Bio-Rollup, a privacy protection solution for biometric recognition systems that combines certificate authority, blockchain Layer 2 scaling, and zero-knowledge proofs. It enhances auditing efficiency through lightweight Merkle proofs, reduces blockchain storage needs, and protects user privacy by preventing unauthorized access and model theft. Experiments on deep neural networks show Bio-Rollup improves system integrity, simplifies deployment, and provides passive defense against data leaks and model-stealing attacks.
[39]	“Security Challenges and Performance Trade-Offs in On-Chain and Off-Chain Blockchain Storage: A Comprehensive Review”	Eren et al.	The study compares on-chain, off-chain, and hybrid blockchain storage, highlighting trade-offs between security, cost, and scalability. Hybrid models balance these factors for optimized storage solutions.
[40]	A Blockchain–Machine Learning Ecosystem for IoT-Based Remote Health Monitoring of Diabetic Patients”	Ratta et al.	The paper presents a blockchain and IoT system for remote diabetes management, using Ethereum smart contracts and machine learning. AdaBoost achieved the highest prediction accuracy of 92.64%, enabling effective patient monitoring and doctor interaction.
[41]	“Analysis of Blockchain-Technology”	Dvorchuk and Shpinareva	The paper gives an overview of blockchain technology; its core features; its challenges, like scalability; and its consensus mechanisms. It reviews Layer 1 and Layer 2 scaling solutions and highlights emerging hybrid consensus and AI-based optimizations.
[42]	“Cost Optimization in Layer 2 Rollups via EIP-4844: A Gas Efficiency and Economic Analysis”	Dyade et al.	The article analyzes Ethereum’s scalability challenges and high fees, focusing on Layer 2 rollups. It evaluates Proto-Danksharding (EIP-4844), which reduces call data costs via blob transactions, improving gas efficiency and making Layer 2 rollups more economically feasible.
[43]	“Blockchain Meets Machine Learning: A Survey”	Khoshgoftaar et al.	The paper surveys the integration of blockchain and machine learning, highlighting benefits like enhanced efficiency, data integrity, and privacy across industries. It also addresses challenges such as security, implementation, data processing, and scalability that need resolution for full potential.
[44]	“Optimizing Academic Certificate Management with Blockchain and Machine Learning: A Novel Approach Using Optimistic Rollups and Fraud Detection”	Tan-Vo et al.	The paper proposes using blockchain with optimistic rollups and machine learning to improve academic certificate management. This approach reduces transaction costs and delays by 61.92% and enhances fraud detection, boosting system security and transparency.
[45]	“Blockchain-Based Electronic Health Passport for Secure Storage and Sharing of Healthcare Data”	Maravi and Mishra	The paper presents a blockchain-based health passport system combining on-chain and off-chain storage with searchable encryption to ensure privacy and efficiency. Enhanced consensus and aggregate signatures improve verification speed and security, offering a robust, scalable solution for managing health data.
[46]	“Survey of Blockchain-Based Applications for IoT”	Enaya et al.	The paper reviews blockchain applications in IoT, emphasizing improved security, scalability, and data integrity. It covers Layer 2 scaling and tokenization and suggests future integration with AI, machine learning, and edge computing for smarter, more secure IoT systems.
[47]	“A dataset of Uniswap daily transaction indices by network”	Chemaya et al.	The article provides a validated dataset of over 50 million daily Uniswap transactions across Ethereum Layer 1 and Layer 2 networks. It addresses gaps in granular Decentralized Finance (DeFi) transaction data, offering daily indices on volume, users, and wealth distribution. The dataset highlights Ethereum’s dominance and growing Layer 2 adoption due to lower fees and faster trades. It supports multidisciplinary research on DeFi scalability, decentralization, and economic dynamics.
[48]	“Blockchain-enabled federated learning with edge analytics for secure and efficient electronic health records management”	S and K R	The paper presents the EPP-BCFL framework, combining blockchain with privacy techniques to improve federated learning for secure and efficient electronic health records management. It reduces computational costs and communication latency, enhances attack resilience, and achieves high accuracy, supporting scalable and trustworthy healthcare data collaboration.
[49]	“ScaleSFL: A Sharding Solution for Blockchain-Based Federated Learning”	Madill et al.	ScaleSFL is a sharding solution for blockchain-based federated learning that improves scalability and security by verifying model updates off-chain. Implemented on Hyperledger Fabric, it demonstrates linear performance gains and efficient, secure validation for scalable federated learning.
[50]	“W3Chain: A Layer2 Blockchain Defeating the Scalability Trilemma”	Xu et al.	W3Chain is a Layer 2 blockchain that overcomes the scalability trilemma by decoupling correctness and using committee reconfiguration and cross-shard handling. It achieves over 10.000 transactions per second (TPS) with low latency, ensuring security and decentralization.
[51]	“Data Centric Blockchain Based Evaluation Approach to Analyze E-Commerce Reviews Using Machine and Deep Learning Techniques”	Acheampong et al.	The paper presents a blockchain and IPFS-based architecture for secure, scalable e-commerce review analysis. Deep-learning models outperform traditional ones, improving accuracy by up to 4% with larger datasets.

presents a compilation of articles included in this systematic review, accompanied by summaries that encapsulate their principal findings and contributions. This overview facilitates a comprehensive mapping of the existing research landscape, providing readers with a concise understanding of the principal themes and innovations addressed within the literature. Section 4 offers an in-depth comparative analysis of these works, emphasizing each study’s key contributions. Furthermore, this section categorizes the articles based on several criteria, such as principal contributions and limitations. This classification seeks to deliver a clear and structured perspective on the current state of Layer 2 blockchain scalability solutions and a possible introduction of machine learning in this area, supporting a refined assessment of progress, existing limitations, and persistent challenges in the domain.

In addition to the summarized overview provided in Table 4, a more structured classification of the included studies is presented in

Table 5. Comparative categorization of included studies highlighting research contributions and employed methodologies.

Research Title	Primary Research Contributions	Employed Methodology
“Trustworthy Reputation for Federated Learning in O-RAN Using Blockchain and Smart Contracts”	Blockchain-based framework for trustworthy federated learning in O-RAN, introducing smart contracts and an on-chain reputation system	Framework design, smart contracts on Polygon Layer 2 and oracle integration, and O-RAN via dApps integration
“A survey of Layer-two blockchain protocols”	Comprehensive taxonomy and comparative analysis of Layer 2 protocols addressing scalability issues	Systematic literature review of Layer 2 mechanisms and classification by type
“A Survey on Data Availability in Layer 2 Blockchain Rollups: Open Challenges and Future Improvements”	Identified data availability as key challenge; reviewed rollup designs for security and efficiency	Literature review of rollup mechanisms with focus on data posting and verification
“Architecture for the Academic Certificate System on the Ethereum Layer 2 Solution”	Proposed Layer 2 architecture for certificate management using Identifiers IPFS and Decentralized Identifiers	System architecture design, technology comparison, prototype implementation
“Assessing the Environmental Sustainability of Polygons Consensus Mechanism and Transaction Processing, Comparing Its Energy Consumption and Carbon Footprint with Other Layer 2 and Layer 1 Blockchain Solutions, And Exploring Potential Avenues for Further Optimization”	Evaluated energy consumption and carbon footprint, highlighting proof-of-stake environmental benefits	Comparative carbon footprint and energy consumption analysis of Polygon and other blockchains
“Designing an Intelligent System for Vaccine Supply Chain Management Based on Blockchain Using Machine Learning Algorithms”	Integrated machine learning and blockchain for scalable, secure vaccine tracking and expiration detection	ML models (Long Short-Term Memory and Support Vector Machines) for credibility and prediction, blockchain sharding for scalability, and IoT data integration
“Bio-Rollup: a new privacy protection solution for biometrics based on two-layer scalability-focused blockchain”	Combined Layer 2 rollups, zero-knowledge proofs, and certificate authorities to enhance privacy and auditing	Blockchain rollups, zero-knowledge proof, and deep neural network experiments
“Security Challenges and Performance Trade-Offs in On-Chain and Off-Chain Blockchain Storage: A Comprehensive Review”	Analyzed on-chain vs. off-chain storage trade-offs with hybrid solutions balancing security and scalability	Comprehensive review comparing security, cost, and scalability implications
“A Blockchain–Machine Learning Ecosystem for IoT-Based Remote Health Monitoring of Diabetic Patients”	Developed blockchain–IoT system using smart contracts and AdaBoost for accurate patient health monitoring	Ethereum smart contracts; AdaBoost machine learning, achieving 92.64% accuracy
“Analysis of Blockchain-Technology”	Overview of blockchain features, scalability challenges, consensus mechanisms, and AI-based optimizations	Literature review with focus on Layer 1 and Layer 2 scaling solutions
“Cost Optimization in Layer 2 Rollups via EIP-4844: A Gas Efficiency and Economic Analysis”	Evaluated Proto-Danksharding for reducing call data cost and gas fees to optimize Layer 2 rollups economic feasibility	Economic and gas cost analysis
“Blockchain Meets Machine Learning: A Survey”	Surveyed blockchain and machine-learning integration benefits and challenges across industries	Review synthesis, cross-domain analysis
“Optimizing Academic Certificate Management with Blockchain and Machine Learning: A Novel Approach Using Optimistic Rollups and Fraud Detection”	Reduced costs and delays in certification via optimistic rollups and fraud detection	System design and implementation, reducing transaction costs by 61.92% and integrating ML fraud detection
“Blockchain-Based Electronic Health Passport for Secure Storage and Sharing of Healthcare Data”	Blockchain-based electronic health passport with on/off-chain storage	System design with aggregate signatures and enhanced consensus
“Survey of Blockchain-Based Applications for IoT”	Overview of blockchain role in IoT, focusing on security, scalability, Layer 2 solutions, and future AI integration	Literature review of blockchain and IoT applications
“A dataset of Uniswap daily transaction indices by network”	Provided large-scale dataset of Layer 1 and Layer 2 DeFi transactions enabling economic and scalability research	Data collection, validation, and publication of DeFi dataset
“Blockchain-enabled federated learning with edge analytics for secure and efficient electronic health records management”	Proposed blockchain with privacy for federated learning improving healthcare data collaboration efficiency	Federated learning framework design and evaluation
“ScaleSFL: A Sharding Solution for Blockchain-Based Federated Learning”	Sharding solution improving scalability and security of federated learning on blockchain	Sharding design on Hyperledger Fabric, experimental measurements
“W3Chain: A Layer2 Blockchain Defeating the Scalability Trilemma”	Layer 2 blockchain that achieves defeating scalability trilemma (security, scalability, decentralization)	Protocol design with committee reconfiguration, cross-shard handling, achieves > 10,000 TPS
“Data Centric Blockchain Based Evaluation Approach to Analyze E-Commerce Reviews Using Machine and Deep Learning Techniques”	Blockchain-based secure framework with enhanced accuracy using deep learning for review analysis	Blockchain and IPFS architecture, ML/deep-learning evaluation

. This table categorizes each article according to its primary research contributions and employed methodology. By organizing the literature along these dimensions, Table 5 complements the descriptive synthesis of Table 4 and enables a clearer comparative evaluation of the state of research.

Table 6. Comparative categorization of included studies highlighting limitations and research gaps.

Research Title	Reported Limitations	Identified Research Gaps
“Trustworthy Reputation for Federated Learning in O-RAN Using Blockchain and Smart Contracts”	Limited scalability evaluation; need for extensive real-world testing	Future scalability improvements; cross-vendor trust robustness
“A survey of Layer-two blockchain protocols”	No experimental evaluation; descriptive analysis only	Dynamic protocol adaptation and optimization strategies
“A Survey on Data Availability in Layer 2 Blockchain Rollups: Open Challenges and Future Improvements”	Data availability remains a bottleneck preventing full security	Mechanisms to improve reliable off-chain data posting
“Architecture for the Academic Certificate System on the Ethereum Layer 2 Solution”	Implementation restricted to academic certificate domain; performance comparison limited	Broader application to other dApps and evaluation under high transaction load
“Assessing the Environmental Sustainability of Polygons Consensus Mechanism and Transaction Processing, Comparing Its Energy Consumption and Carbon Footprint with Other Layer 2 and Layer 1 Blockchain Solutions, And Exploring Potential Avenues for Further Optimization”	Limited analysis on long-term governance and renewable integration	Further renewable energy integration and consensus optimization
“Designing an Intelligent System for Vaccine Supply Chain Management Based on Blockchain Using Machine Learning Algorithms>	Domain limited to vaccines; limited dataset diversity	Scalability testing in diverse global environments
“Bio-Rollup: A New Privacy Protection Solution for Biometrics Based on Two-Layer Scalability-Focused Blockchain”	Storage requirements; deployment complexity	Scalability under a wider adoption of biometric datasets
“Security Challenges and Performance Trade-Offs in On-Chain and Off-Chain Blockchain Storage: A Comprehensive Review”	Storage overhead; cost–security balance challenges	Need for performance-oriented frameworks that balance storage scalability with security guarantees.
“A Blockchain–Machine Learning Ecosystem for IoT-Based Remote Health Monitoring of Diabetic Patients”	Limited to diabetic case; need broader patient data	Larger-scale, real-world healthcare deployment required for validation
“Analysis of Blockchain-Technology”	Lack of experimental results on AI optimizations	Practical implementations of AI-enhanced blockchain scalability
“Cost Optimization in Layer 2 Rollups via EIP-4844: A Gas Efficiency and Economic Analysis”	Early-stage EIP evaluation; real-world deployment data missing	Extensive testing on diverse network conditions
“Blockchain Meets Machine Learning: A Survey”	Security, scalability, and data handling challenges	Addressing implementations hurdles for large-scale adoption
“Optimizing Academic Certificate Management with Blockchain and Machine Learning: A Novel Approach Using Optimistic Rollups and Fraud Detection”	Narrow domain; security performance under adversarial conditions not fully tested.	Expansion to other digital credentials and fraud types
“Blockchain-Based Electronic Health Passport for Secure Storage and Sharing of Healthcare Data”	Scalability under heavy load; privacy-preserving trade-offs	Scalability in diverse healthcare systems
“Survey of Blockchain-Based Applications for IoT”	Fast evolving IoT requirements; integration challenges	AI and edge computing integration in scalable IoT blockchain systems
“A dataset of Uniswap daily transaction indices by network”	Limited to Uniswap; no predictive ML analysis conducted	Expansion to multi-platform datasets and ML-driven DeFi insights
“Blockchain-enabled federated learning with edge analytics for secure and efficient electronic health records management”	Communication latency; computational overhead constraints	Efficient scaling of blockchain–federated learning in healthcare
“ScaleSFL: A Sharding Solution for Blockchain-Based Federated Learning”	Limited cross-shard communication details; tested in controlled environment; scalability at extreme scale not verified	Addressing cross-shard consistency and latency
“W3Chain: A Layer2 Blockchain Defeating the Scalability Trilemma”	Prototype benchmark; not tested under adversarial network conditions	Need for validation under real-world, adversarial workloads
“Data Centric Blockchain Based Evaluation Approach to Analyze E-Commerce Reviews Using Machine and Deep Learning Techniques”	Dataset bias; scalability in high-volume deployments	Expansion to multi-domain reviews and real-time analytics

highlights the reported limitations of each paper alongside the research gaps identified for future exploration. By systematically associating reported limitations with corresponding research gaps, the table provides a structured overview of where current solutions remain insufficient and identifies promising directions for future investigations aimed at developing more resilient, secure, and scalable blockchain-based ecosystems.

5. Research Findings

The analysis of publication trends highlights a growing academic interest in the adoption of blockchain technology. Among the domains concerned with secure information management, Layer 2 solutions have attracted particular attention, as they address the critical scalability limitations of Layer 1 blockchains. Nonetheless, given the importance of the topic and the relatively limited number of studies analyzed, further investigations are required before the field can be regarded as sufficiently mature to gain broad acceptance.

The systematic review of 20 studies contributes comprehensive insights into the three research questions, reinforcing the evidence base and refining the interpretations drawn from individual works.

5.1. RQ1: How Have Scalability Issues Been Addressed in the Ethereum Blockchain?

5.1.1. Scalability Issues

Blockchain technology, initially developed for Bitcoin transactions, has evolved to support diverse applications across various sectors. As blockchain networks grow and include a large number of participating nodes, scalability has become a pressing issue. This challenge stems from the need to replicate the entire distributed ledger on every node in the network, thus increasing the computational burden [22]. Additionally, the block validation process can suffer from reduced transaction throughput, higher storage requirements, increased latency, and greater energy consumption, all of which limit the system’s ability to scale efficiently [36].

Some factors related to blockchain scalability issues can be found in

Table 7. Key factors affecting blockchain scalability.

Factor	Description	Articles
Transaction throughput	This refers to the maximum number of transactions the protocol can process per second.	[14,39,41,46]
Latency	This relates to the time required for a transaction to be initiated and reach consensus, a process also known as finality.	[39,41,46]
Block size	This refers to the total storage capacity of a block allocated for transactions. If a block exceeds this limit, the network will reject it.	[39,41]
Data availability	It represents a challenge where transaction data must remain accessible to all participants in the network.	[39]
Number of nodes	This denotes the total count of nodes present within the blockchain network.	[39,41,46]
Network load	This indicates the volume of transactions handled by the network.	[39,41]
Consensus model	The consensus mechanism refers to the procedure through which blockchain transactions are validated and approved.	[39,41,46]
Computation energy	This reflects whether the algorithm (or the system employing it) requires substantial energy consumption for block mining.	[41,46]
Cost issue	This refers to the overall cost incurred in the process of verifying a transaction within the blockchain.	[39,41,46]
Storage	It denotes the overall storage capacity that a blockchain network is capable of utilizing.	[39,46]

.

The scalability of blockchain systems is influenced by several key factors, including technical design, consensus protocols, and infrastructure limitations, all of which determine the network’s capacity to efficiently manage growing volumes of data and users [52,53].

A central metric in assessing this capacity is transaction throughput, measured in transactions per second (TPS), which reflects the aggregate ability of the entire network to process and record transactions, rather than the performance of individual nodes. Throughput is shaped by various interconnected elements, such as block size, block time, consensus mechanisms, network congestion, transaction complexity, and overall network architecture [48]. Throughput encompasses both read and write operations within the blockchain. Read performance, which involves how efficiently nodes respond to data-retrieval requests, becomes increasingly challenging, as the blockchain grows continuously in an append-only fashion. As the ledger expands, it becomes impractical for most users to store a full copy locally, leading many to rely on lightweight nodes that depend on external full nodes or servers for data access. This dependence places significant strain on full nodes, potentially creating bottlenecks that reduce read throughput—the speed at which data queries are processed. Write throughput, or transaction throughput, is concerned with how many transactions can be validated and included in each block. While increasing the block size could theoretically improve transaction throughput, most blockchain systems enforce strict limits to safeguard security and network stability. For example, Bitcoin restricts block size to 1 MB [54] because larger blocks may cause longer propagation delays, increase security vulnerabilities, and expose the network to denial-of-service attacks [55]. Thus, achieving an optimal block size is critical to balancing transaction throughput with security considerations, ensuring the blockchain maintains both high performance and network integrity.

Reducing the time between the creation of consecutive blocks can also improve throughput, as more transactions are confirmed per unit time [9]. Nonetheless, this approach poses security risks. When blocks are generated too frequently, the likelihood of forks increases due to asynchronous node-processing capabilities. These forks can compromise the network’s integrity, raising the probability of double-spending and other malicious behaviors [56]. As such, a careful balance must be struck between reducing block time and preserving the network’s security.

An influence on throughput and overall scalability is also exerted by the consensus mechanism, which enables decentralized participants to agree on the validity of transactions without relying on a central authority [55]. By addressing trust and conflict resolution within distributed networks, consensus protocols play a critical role in shaping a blockchain’s performance. The development of blockchain technology has led to a diverse range of consensus mechanisms, each designed to balance decentralization, security, and scalability challenges, with

Table 8. Consensus mechanisms.

Consensus Name	Description	Examples of Blockchains That Use It
Proof of work	In proof of work, miners solve complex mathematical puzzles to add new blocks to the blockchain. The first one to solve the puzzle gets to add the block and is rewarded with cryptocurrency.	Bitcoin
Proof of stake	Proof of stake requires participants to lock up a certain amount of cryptocurrency as collateral, with the chance of validating blocks proportional to the amount staked.	Ethereum
Pure proof of stake	Pure proof of stake is a variation of proof of stake where the probability of creating a new block is determined by the amount of stake held by a participant.	Algorand
Secure proof of stake	Secure proof of stake is an optimized version of proof of stake used to enhance security, focusing on preventing attacks and improving scalability.	MultiversX (Elrond)
Proof of history	Proof of history is a mechanism that creates a historical record to prove that an event has occurred at a specific moment in time, reducing the workload in consensus.	Solana
Practical Byzantine Fault Tolerance	Practical Byzantine Fault Tolerance ensures that a transaction is valid if a supermajority of nodes agree on it, even in the presence of faulty or malicious nodes.	Hyperledger Fabric
Ouroboros	Ouroboros is a proof-of-stake-based consensus protocol that aims to secure the blockchain with random selection of validators based on their stake.	Cardano

summarizing the most widely adopted ones [57]. However, Layer 1 scalability solutions often involve fundamental architectural changes, such as modifying consensus algorithms or implementing sharding, that can compromise backward compatibility and lead to issues like network forks [14].

Consensus mechanisms vary significantly in their operational efficiency, with each protocol demonstrating unique performance characteristics, particularly in terms of TPS and block-creation speed, as presented in

Table 9. Consensus mechanisms’ characteristics.

Consensus Name	Transactions per Second	Block-Creation Speed
Proof of work	~7	~10 min
Proof of stake	>30	~12 s
Pure proof of stake	>1000	<5 s
Secure proof of stake	>15,000	~6 s
Proof of history	2000–65,000	~400 milliseconds
Practical Byzantine Fault Tolerance	~1000	~1–2 s
Ouroboros	~1000	~20 s

, based on results from [51,52,58].

The computational energy consumption inherent to a blockchain system is intrinsically linked to the design of its consensus mechanism. Energy-intensive protocols, such as the proof-of-work protocol [59], impose considerable demands on computational resources, thereby constraining scalability by limiting operational efficiency and raising sustainability concerns.

The scalability of a blockchain network is not only influenced by its throughput capacity and consensus design but is also closely linked to how it handles periods of high transactional demand. When the volume of transactions exceeds the network’s ability to process them within a given timeframe, congestion occurs—resulting in slower confirmation times, elevated transaction costs, and reduced efficiency. These effects reflect the limitations of the underlying architecture and highlight the need for scalable solutions. Approaches such as increasing block size, improving consensus protocols, implementing Layer 2 technologies, and strengthening infrastructure are among the most common strategies used to alleviate congestion and support more sustainable transaction processing [60].

Scalability in blockchain systems is also significantly influenced by storage requirements, which grow in parallel with the continuous expansion of the ledger [9]. As new blocks are appended permanently and immutably, the total volume of stored data increases substantially over time. This growing demand places pressure on both individual nodes and the broader network infrastructure, particularly in terms of storage capacity, synchronization speed, and bandwidth consumption. As a result, maintaining a full copy of the blockchain becomes increasingly impractical for ordinary users, pushing many toward reliance on centralized or third-party services to access and verify data. This reliance not only raises concerns about accessibility and performance but also challenges the decentralized ethos that underpins blockchain technology. Therefore, managing storage growth is a crucial component of addressing long-term scalability and preserving the network’s openness and resilience.

The escalating demands placed on blockchain networks underscore the pressing need for novel and effective approaches that can enhance scalability, ensuring the technology remains viable for broader adoption without undermining its foundational attributes.

5.1.2. Scalability Solutions

Existing scalability approaches are generally divided into two main categories: Layer 1 and Layer 2 solutions [36]. Layer 1 solutions aim to enhance scalability by altering core parameters of the main blockchain, such as increasing block size, modifying the consensus algorithm, or implementing network sharding [34]. A notable example is Ethereum’s shift from a proof-of-work protocol to a proof-of-stake protocol, a Layer 1 strategy intended to improve transaction throughput by changing the consensus protocol. However, Layer 1 approaches require fundamental modifications to the blockchain architecture, which can introduce new challenges and may result in network forks [34].

Ethereum, a globally accessible public blockchain, encounters notable limitations due to high transaction congestion and excessive fees. As of 2023, it managed only 20–30 TPS, with average transaction fees around USD 7 and peaking at USD 40 in 2021 [35]. These escalating and unsustainable costs, expected to rise further with increased user adoption, present a significant barrier to the platform’s broader scalability and mainstream use [35].

According to [43], a Layer 1 solution would be sharding, which is a proposed scalability enhancement for the Ethereum network that involves dividing it into smaller, independent units known as shards. Each shard functions as a distinct blockchain with its own validators, transaction records, and smart contracts, enabling parallel transaction processing and significantly increasing overall network throughput. In this architecture, transactions are handled by specific shards rather than the entire network, allowing processing capacity to scale proportionally with the number of shards. While this approach promises to support thousands of transactions per second, it also introduces technical challenges, particularly in maintaining consistency and security across shards through effective cross-shard communication. Ongoing research aims to address these complexities. If successfully implemented, sharding could substantially improve Ethereum’s scalability, making it more capable of supporting high-volume applications such as financial services, supply chain systems, and decentralized exchanges.

Layer 2 solutions decrease the transaction load on the main blockchain, offering a more efficient and scalable infrastructure [14,39,41,42]. These solutions handle data processing on a secondary layer, with only summarized information retained on the main blockchain [39]. The primary Layer 2 scaling solutions include state channels; sidechains; plasma; rollups; or a combination of Layer 2 solutions, called hybrid solutions [34,36,38,41,42,44,46].

Table 10. A comparative analysis of Layer 2 scaling solutions.

Layer 2 Solution	Key Features	Advantages	Challenges
State channels	Versatile off-chain interactions with final settlement conducted on-chain upon completion	Enhanced functionality beyond payment processing and decreased network congestion	Complex configuration process and restricted to designated participants; reduced decentralization, requiring online counterparties
Sidechains	Adjustable consensus methods and functions separately from the main blockchain	Enables scalability and innovation, suitable for large-scale applications	Needs its own security model and may have limited decentralization
Plasma	Child chains linked to Layer 1	Faster transaction processing	Exit fraud risks, complex withdrawals
Rollups	Process transactions on Layer 2 and store summary data on Layer 1	High security, fast verification	Fraud-proof verification delay (up to a week)
Hybrid solutions	Integrates elements of various Layer 2 solutions, offering a balance between scalability and security	Offers high scalability and adaptability for a wide range of use cases	Greater design complexity and reliance on synchronization between layers

shows the comparative analysis of Layer 2 scaling solutions, examining their key features, advantages, and challenges.

5.2. RQ2: What Are the Key Types of Ethereum Layer 2 Scaling Solutions Currently Implemented, and How Do They Differ in Terms of Architecture, Security, and Scalability?

Based on the survey in [14], the principal types of Ethereum Layer 2 scaling solutions currently implemented include state channels, sidechains, plasma chains, rollups (optimistic and zero-knowledge rollups), and hybrid approaches. These solutions differ significantly in architecture, security models, and scalability potential. In addition, there is also a more advanced solution, called validium [34]. Understanding these scaling mechanisms provides the foundation for examining how they support dApps, which increasingly depend on Layer 2 technologies for enhanced performance.

Decentralized applications are blockchain-based applications that operate on a distributed network rather than relying on a centralized server [18]. They serve a variety of functions across multiple domains, with Layer 2 technologies playing a crucial role in enhancing their efficiency and scalability. Figure 11 illustrates these principal Layer 2 solutions, highlighting their core operational features and interactions with the underlying Layer 1 blockchain [14,18]. Each node highlights the key characteristics, including transaction-handling method, interaction with the main chain, security considerations, and scalability advantages. While many of these technologies were initially designed for Ethereum, their underlying architectures and principles are broadly applicable to other blockchain platforms [18].

5.2.1. State Channels

State channels are an off-chain solution designed to facilitate micro-payments by conducting transactions outside the main blockchain through a secure, pre-established channel [34]. These transactions are ultimately recorded on the main chain as a single consolidated transaction once the payment process concludes. Since state channel transactions occur off-chain, they rely on mutual trust, agreement, and secure communication between participants [34]. A key risk arises if a participant goes offline or fails to close the channel by submitting the final state to the blockchain within a designated timeframe. In such cases, a malicious party could exploit the system by submitting an outdated state or withholding the final state, leading to incorrect fund settlement. Consequently, this approach depends on trusted interactions and secure communication channels among all involved parties [14]. The typical lifecycle of a state channel comprises three phases: establishment, execution, and termination. During the establishment phase, participants initially lock a certain amount of funds or assets in a smart contract on the main blockchain. The total amount of these locked funds defines the capacity of the state channel [14]. While state channels enhance scalability by keeping transactions off-chain within secure channels, sidechains adopt a distinct strategy by establishing independent blockchains that interact with the main chain.

5.2.2. Sidechains

Sidechains involve the deployment of a secondary blockchain that operates as an extension of the primary (main) chain, enabling asset transfers between the main and secondary chains at predefined rates [34]. While the secondary chain’s existence is dependent on the main chain, both chains function independently and may employ distinct consensus mechanisms [14]. Consequently, security vulnerabilities are not shared between the two chains. The integrity and security of a sidechain solution rely entirely on the design and robustness of the secondary chain, as the Layer 1 protocol does not provide inherent security guarantees for the sidechain 46]. In contrast to sidechains, which function as parallel blockchains with their own consensus rules, plasma introduces a hierarchical structure of subchains anchored to the main chain, designed to improve scalability through layered processing.

5.2.3. Plasma

Plasma operates by processing data on subchains while recording transaction summaries on the main blockchain. This approach offers a secure Layer 2 solution; however, coordination among subchains presents operational challenges and adds complexity to the system [39]. Plasma chains rely on an optimistic model in which transactions are presumed valid unless a designated participant, known as a watcher, detects malicious activity and initiates a dispute; the network then determines whether the fault lies with the publisher of the transaction or the reporting watcher and imposes penalties accordingly. Although this model offers high efficiency under normal conditions, it necessitates a predefined dispute window during which watchers can challenge fraudulent transactions; this dispute period inherently delays the finalization of transactions [17].

As limitations, plasma chains are vulnerable to exit fraud and involve complex withdrawal processes, which limit their practicality for real-time applications [39]. Although plasma reduces the data load on the main blockchain, it faces challenges with withdrawal delays and fraud proofs. Rollups address these limitations by aggregating transactions off-chain and posting concise proofs on-chain, offering a more efficient and practical scaling method.

5.2.4. Rollups

Rollups execute transactions on Layer 2 while storing summary data on Layer 1, offering substantial scalability benefits, especially for networks handling large volumes of transactions; however, their adoption may be limited by technical complexity and implementation challenges [39].

Rollups are similar to plasma, as transaction execution is shifted to Layer 2, but all execution data is published back on the main chain, allowing the Layer 1 blockchain to record transactions processed by Layer 2. Introduced by Barry Whitehat in 2018 [61], rollups have since become a robust solution to scalability challenges, enabling off-chain transaction processing while preserving the security and decentralization of the underlying Layer 1 blockchain [34]. There are two main types of rollup-based Layer 2 solutions, ZK rollups and optimistic rollups. ZK rollups operate off-chain using zero-knowledge proofs to aggregate multiple transactions into a single lightweight transaction that is posted back to the main chain. In contrast, optimistic rollups do not require zero-knowledge proofs, reducing computational overhead but introducing additional trust assumptions [38].

Optimistic rollups adopt an optimistic approach by presuming transactions to be valid unless proven otherwise, thereby avoiding default verification computations and significantly enhancing scalability [14,34]. However, the smart contract records a history of state root updates along with associated batch hashes. If a batch is disputed, a fraud proof must be submitted on-chain, which the contract verifies. If the proof is valid, the contract reverts the incorrect batch, along with all subsequent batches [14,34]. This approach significantly reduces on-chain computation and transaction verification on Layer 1, enhancing scalability. However, it introduces an unavoidable challenge period during which transactions may be contested, potentially lasting up to one week. As a result, delayed transaction finality becomes a key limitation of optimistic rollups, posing challenges for applications that require immediate settlement. Additionally, the security of optimistic rollups depends on an actively engaged network of nodes capable of identifying and disputing fraudulent transactions, making the system heavily reliant on network integrity and the honesty of participants. As illustrated in Figure 1, fraud proofs are handled on the Layer 2 blockchain, while only the aggregated transaction data is submitted to Layer 1 [14,34].

ZK rollups use ZK proofs to boost scalability and privacy by validating transactions off-chain without revealing their content. Execution results are compressed into a batch and submitted to Layer 1 along with a cryptographic proof, which an on-chain smart contract verifies to finalize transactions instantly, without a challenge period [14,34]. Zero-knowledge proofs are a fundamental component of ZK rollups. In cryptography, they enable a prover to demonstrate the truth of a statement to a verifier without revealing any additional information [34]. These proofs are characterized by three main properties: soundness, which ensures that false claims cannot be accepted; completeness, which guarantees that true claims can be verified; and zero knowledge, which means no information beyond the truth of the statement is disclosed [34].

While ZK rollups leverage zero-knowledge proofs to optimize scalability and privacy, hybrid Layer 2 solutions offer dApps greater flexibility by combining multiple technologies to meet specific functional requirements. This approach enables platforms to balance throughput, cost efficiency, and compatibility with existing smart contracts, as illustrated by implementations such as Arbitrum, TrueBit, and Polygon.

A brief comparison of ZK rollups and optimistic rollups is also provided in

Table 11. A comparative analysis of rollups solutions.

Feature	Optimistic Rollups	ZK Rollups
Verification method	Uses fraud proofs that involve a challenge period to dispute transactions.	Employs zero-knowledge proofs to validate transactions, enabling immediate verification without the need for challenges.
Data handling	Stores complete transaction data on-chain to enable possible fraud proofs and challenges.	Records only transaction proofs and state changes on-chain, minimizing data load.
Withdrawal delay	Withdrawals may take up to one week because of the mandatory fraud-proof period.	Allows instant withdrawals because transactions achieve immediate finality.
Transaction finality	Transaction finality is delayed until the fraud-proof challenge period expires.	Transaction finality is immediate since cryptographic proofs verify validity upon transaction submission.
Security assumption	Assumes that validators act honestly and will promptly challenge invalid transactions during the dispute period.	Depends on the mathematical soundness of zero-knowledge proofs, assuming the proofs are generated correctly.

.

5.2.5. Hybrid Solutions

Hybrid Layer 2 solutions provide decentralized applications with the flexibility to adopt technologies best aligned with their specific requirements [14,18]. Arbitrum, which employs optimistic rollups, enhances transaction throughput and minimizes gas fees while maintaining compatibility with Ethereum smart contracts and supporting existing dApps such as GMX, Radiant Capital, and Camelot; it also accommodates non-DeFi initiatives like Treasure DAO, a casual gaming project focused on creativity [18]. TrueBit addresses computational bottlenecks by delegating resource-intensive tasks off-chain, thus reducing blockchain load and transaction expenses, it employs Interactive Verification, where external nodes validate the correctness of off-chain computations through a gamified challenge-response protocol, and is being explored for use cases in projects like Golem, Dogethereum, and Livepeer, a decentralized video streaming network [18]. Polygon incorporates a range of Layer 2 technologies, including plasma, state channels, ZK rollups, and optimistic rollups, enabling support for a wide array of dApps such as Aave Protocol, Decentral Games, Sandbox, and Uniswap v2, all of which benefit from Polygon’s scalable infrastructure and low-cost environment [18]. The selection of Layer 2 solutions by dApps is driven by functional needs, with platforms like Polygon and Optimism preferred for high-throughput, cost-efficient DeFi and NFT applications, while others, such as Axie Infinity, may adopt custom sidechains to support specialized features and greater control [18]. While hybrid solutions combine multiple Layer 2 technologies to meet diverse application needs, validiums represent a specialized advancement of rollup architectures, focusing on off-chain data storage and controlled data availability to further enhance scalability, albeit with a trade-off in trust assumptions.

5.2.6. Validiums

In [34], it is stated that validiums are advanced systems comparable to rollups but differ in their data-availability method. Unlike rollups, validiums keep transaction data and the state root hash off the main chain, while validity proofs remain on-chain. The accessibility of off-chain data is controlled by a data-availability committee, typically a centralized oracle, which requires a degree of trust [34].

5.3. RQ3: How Can Machine-Learning Techniques Be Integrated with Layer 2 Solutions to Further Optimize Blockchain Scalability and Performance?

Having examined the various Layer 2 scaling solutions and their relative advantages, the next step is to explore how emerging techniques, such as machine learning, can be integrated into these frameworks. By leveraging predictive analytics, anomaly detection, and optimization algorithms, machine learning can enhance transaction efficiency, resource allocation, and overall blockchain performance.

Machine learning, a key subfield of artificial intelligence, replicates the learning processes of the human brain [43]. Its rapid advancement in recent years has resulted in numerous applications that enhance everyday life. Machine learning focuses on developing algorithms and models capable of analyzing and learning from data to make predictions or decisions without explicit programming [43]. This enables systems to improve over time and adapt to different tasks, including image and speech recognition, recommendation engines, and natural language processing [43].

The integration of blockchain and machine learning offers a secure, decentralized, and efficient framework for network transactions and system administration. It enhances data privacy, model reliability, and decision-making through decentralized ownership and access control. This synergy supports scalable and transparent data storage for machine-learning training. It also enables collaborative, privacy-preserving, and decentralized machine-learning ecosystems [37,43,44]. Building on this synergy between blockchain and machine learning, recent studies have explored its application in securing decentralized IoT systems, addressing challenges such as privacy, trust, and efficient data management.

In [62], the authors present the predominance of research focusing on specific vulnerabilities in decentralized IoT applications, noting multiple solutions addressing privacy and trust concerns. To overcome these challenges, blockchain-based authentication frameworks are increasingly employed to decentralize identity verification and secure device interactions. However, existing systems face issues of complexity and storage overhead. The article proposes a novel permissioned blockchain approach optimized for large-scale IoT, incorporating efficient data storage and lightweight authentication, and pioneering the use of homomorphic encryption for user-end data security before cloud upload. Evaluations demonstrate that this trust-centric framework markedly improves privacy and security in IoT environments. Beyond IoT applications, machine-learning techniques are also increasingly applied to cybersecurity, providing advanced methods for threat detection, smart contract analysis, and vulnerability mitigation in Layer 2 blockchain solutions.

In [63], recent research on cybersecurity incidents is reviewed, emphasizing the role of machine-learning techniques in enhancing security measures. It systematically analyzes prior studies to demonstrate how various machine-learning algorithms can be applied to detect and mitigate cyberattacks, including novel threats. The findings indicate that machine learning substantially advances cybersecurity by improving threat identification and response. Additionally, the study maps key machine-learning methods to specific types of cyber threats, evaluating their effectiveness in addressing diverse security challenges. This synthesis underscores the promising impact of machine learning in fortifying digital defense systems. Proposed techniques such as supervised classification, unsupervised anomaly detection, reinforcement learning, and vulnerability prediction play crucial roles. These methods provide tools for detecting cyber threats; assessing smart contract safety; and uncovering exploits within Layer 2 solutions, like rollups, plasma, and sidechains. The review suggests deploying both static and dynamic machine learning-based code analysis to proactively identify smart contract vulnerabilities, enhancing the security and dependability of scalable blockchain implementations. To contextualize these cybersecurity applications, it is useful to review the main categories of machine-learning and AI techniques that are commonly employed to optimize blockchain performance, enhance security, and address scalability challenges.

As illustrated in [11,54], machine learning is commonly divided into four main types: supervised learning, which uses labeled data to train models for prediction; unsupervised learning, which identifies patterns and groups in unlabeled data, semi-supervised learning, which combines a small set of labeled data with a larger unlabeled dataset to improve performance and reduce labeling costs and reinforcement learning, where an agent learns optimal actions by interacting with its environment through rewards in order to learn the best strategies for making decisions. Building on this framework, Yan et al. demonstrates in [12] a comprehensive integration of artificial-intelligence techniques to enhance blockchain performance, notably addressing scalability, security, and privacy concerns. AI methods, including machine learning and deep learning, are applied to optimize consensus algorithms, thereby increasing transaction throughput and reducing latency. Simultaneously, AI facilitates advanced detection of smart contract vulnerabilities, protecting blockchain ecosystems from emerging threats. Privacy-preserving techniques such as FL and differential privacy are also explored, ensuring data confidentiality without compromising security. This AI-driven approach exemplifies how blockchain’s technical limitations can be mitigated through adaptive and intelligent optimization frameworks [12]. These AI-driven techniques are not just theoretical; they have practical applications in enhancing Layer 2 blockchain solutions, as demonstrated in real-world implementations.

In [44], it is demonstrated how optimistic rollups enhance scalability in blockchain-based academic certificate management by enabling off-chain transaction processing that reduces latency and fees. Furthermore, it integrates machine learning-driven fraud detection to identify potentially fraudulent users before transactions finalize, significantly improving security. The machine-learning models achieve high detection accuracy, underpinning proactive defense during the fraud-proof window inherent in optimistic rollups. This combination addresses the challenges of transaction efficiency and timely fraud identification, illustrating the practical synergy of Layer 2 scaling with predictive analytics for secure credential verification systems. Beyond credential verification, machine learning can also optimize security and efficiency in other Layer 2 contexts, such as asset and dispute management.

Focusing on secure asset management, Alyounis and Yasin [64] use machine learning-based predictive analytics and graph-based learning to manage fraud and disputes in blockchain-backed land record systems. The proposed machine-learning models assist in forecasting dispute timelines and detecting anomalous transactions by analyzing relational data structures inherent to blockchains. This could enhance the usability and trustworthiness of plasma chain implementations or similar frameworks where off-chain child chains handle large volumes of transactions. To further address scalability and decentralization challenges in Layer 2 solutions, recent research explores the integration of federated-learning and privacy-preserving machine-learning frameworks.

Javed et al. proposes in [33] a FL framework trusted by blockchain-based reputation and smart contracts that aligns well with Layer 2 concepts that seek scalability and decentralization. In solving sequencer centralization and censorship risks in rollups, FL combined with reputation modeling can decentralize control and ensure reliable participation through dynamic trust scoring. This approach reduces reliance on centralized sequencers while enhancing security and scalability. The blockchain records model updates transparently, and the machine-learning consensus committees selected via FL dynamically optimize communication overhead, indicating a novel synergy between Layer 2 consensus optimization and trustworthy, scalable AI methods. Zero-Knowledge Machine Learning (ZK-ML) ensures that each system component executes algorithms correctly without exposing private data or intermediate states. In this context, ZK-ML principles enhance federated learning by enabling decentralized, privacy-preserving model training without direct data exchange, while providing consistent execution and stronger protection against security threats. The examples above illustrate the diverse ways machine learning can enhance Layer 2 protocols.

Table 12. Conceptual mapping of Ethereum Layer 2 scalability challenges to corresponding machine-learning techniques that enhance performance, trust, and decentralization across various Layer 2 protocols.

Scalability Challenge	Layer 2 Solutions	Relevant Machine-Learning Techniques	Proposed Machine-Learning Integration	Source Studies
Fraud proof, latency, and detection delays	Optimistic rollups	Anomaly detection, predictive modeling	Use anomaly detection models on transaction patterns to anticipate fraudulent behavior early during the fraud-proof challenge period, reducing reaction time. Predictive models could forecast fraud occurrences to prioritize evidence verification.	[44]
High computational cost of cryptographic proof generation	ZK rollups	Reinforcement learning, resource optimization models	Reinforcement-learning agents can optimize parameter tuning and scheduling for proof generation tasks, minimizing computational overhead. Resource-aware machine-learning models may dynamically allocate computation to balance efficiency and security.	[44]
Sequencer centralization and censorship risk	Optimistic and ZK rollups	FL, trust and reputation modeling, ZK-ML	Employ FL across sequencer nodes to decentralize control, while reputation models can assign trust scores to sequencers, enabling dynamic selection of trustworthy entities, reducing centralization and censorship potentials. ZK-ML principles complement FL by enabling decentralized, privacy-preserving model training without direct data sharing, while ensuring consistent execution and improved security against threats.	[33]
Complex exit and dispute resolution processes	Plasma chains	Predictive analytics, Graph Neural Networks (GNNs)	Predictive analytics could forecast exit timeouts and user intent, while GNNs can analyze transaction graphs for dispute patterns, automating alerting and resolution suggestions to enhance usability.	[64]
Real-time authentication and access control under resource constraints	Sidechains, multi-layer blockchains	Lightweight machine learning (e.g., Decision Trees)	Deploy lightweight, incremental machine-learning models at edge nodes for real-time authentication decisions under IoT constraints to accommodate new threats without heavy resource usage.	[62]
Smart contract vulnerabilities exploited via Layer 2	Rollups, plasma, sidechains	Static/dynamic code analysis via machine learning, vulnerability prediction	Machine-learning models trained on contract codebases could predict potential vulnerabilities and exploit patterns, enabling proactive patching and enhanced security in Layer 2 smart contracts.	[63]
Optimizing blockchain scalability, security, and privacy	All Layer 2 and blockchain layers	AI-driven consensus optimization, privacy-preserving ML (e.g., FL and differential privacy)	AI algorithms optimize consensus protocols for efficiency and security while enhancing privacy through techniques like FL and differential privacy, balancing throughput with robust security and confidentiality.	[12]

synthesizes these insights into a conceptual framework, mapping specific scalability and security challenges to corresponding machine-learning techniques.

Table 12 presents a conceptual framework that links specific scalability and security challenges encountered by Ethereum Layer 2 blockchain solutions with corresponding machine-learning techniques that could potentially address these limitations. This synthesis is based on recent scholarly studies and identifies how different Layer 2 protocols, including optimistic rollups, zero-knowledge rollups, plasma chains, sidechains, and state channels, face unique challenges, such as delays in fraud detection, high computational demands, complex dispute resolution, interoperability issues, and susceptibility to malicious actors. The table aligns these challenges with machine-learning approaches such as anomaly detection, predictive modeling, FL, reinforcement learning, and graph-based methods, illustrating how these techniques can improve throughput, security, trust, and decentralization. This analysis highlights promising directions for future research aimed at integrating machine-learning methodologies with Layer 2 blockchain technologies to optimize scalability.

6. Discussions

This review reveals that Ethereum’s network scalability and security challenges have spurred a rich landscape of Layer 2 solutions, each with distinct technical architectures and operational trade-offs. Protocols such as optimistic rollups, zero-knowledge rollups, plasma chains, sidechains, and state channels have been implemented to mitigate congestion, high fees, and latency inherent to Layer 1 transaction processing. While these solutions significantly improve transaction throughput, they introduce new challenges, such as delayed fraud detection, computational overhead, complex cross-chain operations, and increased risk of centralization within off-chain components [14,18,33,43,44,52,55].

Based on the analysis results,

Table 13. Comparison of Layer 2 blockchain solutions.

Solution	Throughput (TPS)	Latency (s)	Security Level	Trust Assumptions
State channels	Very high (1000–10,000)	Low (0.01–0.10)	High	Requires trust in participants
Sidechains	High (100–1000)	Moderate (10–60)	Varies	Depends on sidechain’s security mechanism
Plasma	High (500–5000)	Low to moderate (0.1–1)	High	Trust in plasma operator for data availability
Optimistic rollups	Moderate to high (4000–10,000)	High (0.1–7 days due to fraud-proof period)	High	Trust in fraud-proof mechanism
Zk rollups	High (2000–4500)	Low (0.1–1)	Very High	No additional trust assumptions
Validiums	Very high (9000)	Low (0.1)	High	Trust in data-availability committee

compares the four main kinds of Layer 2 solutions in terms of transaction throughput, latency, security, and trust assumption.

While the table highlights the technical trade-offs of Layer 2 solutions, emerging research suggests that machine-learning techniques can help mitigate some of these limitations.

A core theme across the literature is the emerging role of machine learning to address these persistent challenges. Techniques such as anomaly detection and predictive modeling have been proposed to expedite fraud prevention in rollup protocols, while graph-based models enhance dispute resolution and detection of anomalous behaviors in protocols like plasma. Furthermore, lightweight and federated machine-learning methods are enabling secure, real-time authentication and trust management in constrained IoT and cross-chain environments [14,18,33,43,44,52,55]. These methods are not merely theoretical; several real-world applications illustrate how AI and blockchain integration can enhance performance, security, and decision-making across industries.

Several real-world applications illustrate the integration of blockchain and machine learning. Blackbox AI [65] is an AI coding assistant that streamlines software development by providing real-time code completion, documentation, and debugging support. DHL’s Global Trade Barometer [66] serves as an early indicator of global trade trends by analyzing large volumes of logistics data using AI. IPwe [67], the world’s first global patent register powered by AI and blockchain, leverages natural language processing, predictive analytics, and IBM Watson’s machine learning to analyze patent data, improve transparency, and identify business opportunities. Despite these successes, no single Layer 2 protocol or machine-learning approach fully addresses all challenges, indicating the need for hybrid and adaptive solutions tailored to specific application contexts.

Comparative analysis of approaches suggests that no single Layer 2 protocol or machine-learning method universally solves all scalability and security concerns. Instead, hybrid frameworks, combining multiple protocols and tailored machine-learning algorithms show the most promise for adapting to varied application needs across industries. Building on these insights, further research is necessary to develop comprehensive frameworks that integrate Layer 2 blockchain scalability with advanced machine-learning techniques, particularly for complex, data-intensive environments like IoT.

7. Conclusions

In this systematic review, we were able to identify the significant potential of combining Ethereum Layer 2 solutions with machine learning to overcome key limitations in blockchain technology, such as scalability, high transaction costs, and privacy concerns. Layer 2 protocols, such as rollups, sidechains, and state channels, effectively enhance transaction throughput by moving computations off the main Ethereum chain, while retaining security through on-chain verification. When combined with machine learning, these solutions enable automated trust mechanisms, improved fraud detection, and intelligent data processing within decentralized applications. This collaboration facilitates more transparent, secure, and scalable systems across various domains, including healthcare, supply chain management, and academic certification.

Nonetheless, critical challenges persist, particularly regarding data availability, off-chain computation reliability, and the trade-offs between security and scalability. Despite growing academic interest, significant gaps remain in designing comprehensive frameworks that integrate Layer 2 scalability techniques with real-time machine-learning applications. A particularly promising area for the development and application of the proposed framework is the IoT ecosystem, where the integration of Layer 2 blockchain solutions with machine learning can address critical challenges such as scalability, security, and decentralized decision-making across vast networks of interconnected devices.