Editorial: Machine Learning and Statistical Learning with Applications 2025

1. Introduction

Machine learning and statistical learning have become central to modern scientific discovery and technological innovation. The rapid growth of data across diverse domains has driven the development of increasingly sophisticated models, capable of extracting meaningful patterns from complex, high-dimensional, and often noisy datasets. Recent advances in deep learning, foundation models, and large-scale data-driven approaches have significantly expanded the capabilities of artificial intelligence systems across a wide range of applications [1,2,3,4]. At the same time, the field is evolving beyond a sole focus on predictive performance. Issues such as interpretability, robustness, uncertainty quantification, and data efficiency have become equally important, particularly in high-stakes applications [5,6,7]. This shift reflects a broader movement toward trustworthy and application-aware machine learning systems.

The Special Issue “Machine Learning and Statistical Learning with Applications 2025”, published in the journal Computers (ISSN 2073-431X, MDPI), was conceived to capture these developments by bringing together contributions that span both methodological innovation and real-world applications. The 14 papers included in this issue illustrate the diversity of the field, covering topics such as natural language processing, computer vision, time-series forecasting, scientific computing, remote sensing, and socio-technical systems.

This editorial provides a structured overview of the 14 published articles. Section 2 summarizes each paper’s contributions, organized thematically to highlight the breadth of research covered. Section 3 concludes with reflections on the significance of this collection and directions for future research.

2. An Overview of Published Articles

The papers in this Special Issue can be broadly organized into five thematic areas: healthcare and biomedical analytics, agriculture and environmental applications, forecasting and risk prediction in complex systems, methodological advances in machine learning, and social and enterprise applications. These groupings not only reflect the diversity of application domains but also reveal common methodological trends and shared challenges.

2.1. Healthcare and Biomedical Analytics

Healthcare remains one of the most impactful domains for machine learning, where accurate and interpretable models can significantly improve diagnosis and patient outcomes.

Zhang et al. investigate the classification of textual medical notes from electronic health records using traditional machine learning methods. Their findings demonstrate that well-established models, particularly Logistic Regression, remain competitive when combined with rigorous preprocessing and evaluation strategies, highlighting the continued importance of simplicity and interpretability in clinical contexts.

Hadi et al. employ a DenseNet-based deep learning architecture to detect cardiovascular disease from medical imaging data. Their approach achieves high predictive performance while also incorporating visualization techniques to identify pathological regions, thereby enhancing model interpretability.

Together, these studies illustrate complementary approaches to healthcare analytics, encompassing both text-based and image-based data, and underscore the growing integration of machine learning into clinical workflows.

2.2. Agriculture and Environmental Applications

Machine learning is increasingly applied to agricultural and environmental systems, where it supports monitoring, sustainability, and operational efficiency.

Ljubobratović et al. explore neural network models for predicting peach maturity using tabular data, demonstrating that modern architectures such as TabNet can effectively handle small datasets and enable non-destructive quality assessment in horticulture.

Ad ao et al. address the challenge of limited labeled data by proposing synthetic data generation techniques for deep learning in precision viticulture. By combining rule-based image generation and diffusion models, their approach accelerates model development and reduces reliance on labor-intensive data collection.

Hernández-Macià et al. evaluate machine learning approaches for retrieving Arctic sea ice thickness from satellite radiometry data. Their results show that simpler models, such as Random Forest, can perform comparably to more complex deep learning methods, emphasizing the importance of domain knowledge and physical modeling in environmental applications.

Collectively, these works highlight the importance of practical considerations, such as data availability, interpretability, and deployment constraints, in the successful application of machine learning to real-world environmental and agricultural problems.

2.3. Forecasting and Risk Prediction in Complex Systems

A significant portion of the contributions focuses on forecasting and risk prediction in complex, dynamic, and uncertain systems.

Guritanu et al. introduce a topological machine learning framework based on persistent homology for detecting financial crises, providing interpretable early warning signals from multivariate time-series data.

Yuan et al. propose a hybrid xLSTM–XGBoost model for gasoline price forecasting, demonstrating the effectiveness of combining decomposition techniques with ensemble learning to handle non-stationary and highly volatile data.

In the context of disaster and environmental risk, He and Hu apply machine learning to predict household relocation decisions following disasters, offering insights for policy-making and recovery planning. Satish et al. develop predictive models for tsunami occurrence using seismic and geospatial data, highlighting the effectiveness of Random Forest in handling imbalanced datasets. Leon-Gomez et al. further contribute to environmental forecasting by proposing a UMAP-based data augmentation framework for improving wind speed prediction across multiple datasets.

These studies demonstrate how machine learning can support decision making under uncertainty across a wide range of domains, including finance, energy, disaster management, and environmental monitoring.

2.4. Methodological Advances in Learning, Optimization, and Uncertainty

Several papers focus on advancing the methodological foundations of machine learning. Malashin presents a framework for generating natural-language explanations for static-analysis warnings using transformer models combined with multi-objective optimization. This work contributes to explainable AI by addressing the trade-off between faithfulness and linguistic quality.

Montanari et al. investigate machine learning techniques for uncertainty estimation in dynamic aperture prediction within particle accelerators. By comparing approaches such as Monte Carlo dropout and bootstrap methods, the study highlights the importance of reliable uncertainty quantification in scientific applications.

These contributions reflect a broader shift in the field toward developing machine learning systems that are not only accurate but also interpretable, reliable, and capable of supporting informed decision making in complex settings.

2.5. Social, Political, and Enterprise Applications

The final group of papers demonstrates the application of machine learning to socio-technical and organizational systems.

Sernani et al. analyze digital political campaigning during the 2024 European Union elections using machine learning techniques applied to social media data. Their findings highlight the roles of micro-targeting, algorithmic bias, and disinformation in shaping political outcomes, contributing to ongoing discussions on digital governance and regulation.

Bhaskaran introduces EnterpriseAI, a transformer-based framework for optimizing enterprise systems and reducing operational costs. By leveraging transfer learning and complex dependency modeling, the proposed system enhances decision making and resource allocation in organizational contexts.

These studies underscore the expanding role of machine learning in understanding and optimizing human-centered systems, where technical performance must be balanced with ethical, social, and organizational considerations.

2.6. Cross-Cutting Trends

Across all thematic areas, several important trends emerge. First, many contributions adopt hybrid modeling approaches, combining multiple algorithms to leverage their complementary strengths. Second, there is a strong emphasis on practical challenges, including data scarcity, class imbalance, and deployment constraints. Third, the importance of interpretability and uncertainty estimation is increasingly recognized, particularly in high-stakes applications.

Overall, the papers in this Special Issue demonstrate that machine learning and statistical learning are advancing along both methodological and application-driven dimensions, reinforcing their role as essential tools in modern science and engineering.

3. Conclusions

This Special Issue provides a comprehensive overview of current research in machine learning and statistical learning, highlighting both theoretical developments and practical applications across a wide range of domains. The included papers demonstrate how advanced learning techniques can address complex real-world challenges while also emphasizing the importance of interpretability, robustness, and efficiency.

By organizing the contributions into thematic groups, this editorial has aimed to highlight the connections between different research directions and to identify common trends shaping the field. We hope that this collection will serve as a valuable resource for researchers and practitioners, fostering further innovation and interdisciplinary collaboration.

As Guest Editor, I would like to express my sincere gratitude to all authors for their high-quality contributions, to the reviewers for their careful and constructive evaluations, and to the editorial team of Computers for their support throughout the publication process.