Enhancing Industrial Cyber Security, Focusing on Formulating a Practical Strategy for Making Predictions through Machine Learning Tools in Cloud Computing Environment

Abstract

Cloud computing has revolutionized how industries store, process, and access data. However, the increasing adoption of cloud technology has also raised concerns regarding data security. Machine learning (ML) is a promising technique to enhance cloud computing security. This paper focuses on utilizing ML techniques (Support Vector Machine, XGBoost, and Artificial Neural Networks) to progress cloud computing security in the industry. The selection of 11 important features for the ML study satisfies the study’s objectives. This study identifies gaps in utilizing ML techniques in cloud cyber security. Moreover, this study aims at developing a practical strategy for predicting the employment of machine learning in an Industrial Cloud environment regarding trust and privacy issues. The efficiency of the employed models is assessed by applying validation matrices of precision, accuracy, and recall values, as well as F1 scores, R.O.C. curves, and confusion matrices. The results demonstrated that the X.G.B. model outperformed, in terms of all the matrices, with an accuracy of 97.50%, a precision of 97.60%, a recall value of 97.60%, and an F1 score of 97.50%. This research highlights the potential of ML algorithms in enhancing cloud computing security for industries. It emphasizes the need for continued research and development to create more advanced and efficient security solutions for cloud computing.

1. Introduction

Modern organizations have embraced cloud computing as a necessary component because it allows them to store, access, and analyze data effectively [1]. However, data security concerns also arise with the continuous use of cloud technology [2,3]. Industries are a major target for cyber-attacks because they store tremendous amounts of confidential data on cloud servers, including financial information, client data, and scientific property [4,5]. Therefore, it is essential to enhance cloud computing security to save valuable data from unauthorized access and theft. ML has been recognized as a promising method for enhancing cloud computing security [6]. Different algorithms can examine huge data capacities, identify arrangements, and learn from the data to identify potential security threats. By using machine learning, industries can enhance their security measures and reduce the risk of data breaches [7,8]. One of the significant challenges of securing cloud environments is the lack of control over the physical infrastructure [9,10]. The physical infrastructure, which consists of servers, storage units, and network devices, must be kept up by the cloud package providers. Therefore, to ensure that the data kept on cloud servers are secure, trust between the consumer and the cloud service provider is crucial [11,12]. Another challenge of securing cloud environments is the complexity of the system. Cloud environments involve multiple layers of hardware, software, and networking components, making identifying and addressing security threats difficult [12]. Cloud computing has become increasingly popular for its scalability, flexibility, and cost-saving benefits. However, it also presents several vulnerabilities that can threaten the security and integrity of data and applications [13]. Some major vulnerabilities in cloud computing include data breaches [14], insider threats, malware attacks [15], denial of service (DoS) attacks [16], insecure APIs, shared infrastructure risks, lack of visibility and control, compliance and legal risks, data loss and data leakage, and lack of encryption [17]. Sensitive data kept in the cloud can be stolen by unauthorized parties frequently due to poor access controls, ineffective authentication procedures, or defects in the infrastructure or software of cloud service providers (C.S.P.s) [18]. Insider threats are risks posed by staff members or authorized users with access to protected cloud assets [3]. Malware attacks, including viruses, worms, and ransomware, can infect cloud systems and compromise data integrity and confidentiality [19]. DoS attacks aim to degrade cloud services by overwhelming the cloud infrastructure or applications with excessive traffic or requests [20]. Shared infrastructure risks arise from cloud services built on shared infrastructure, where multiple users and applications share the same physical resources [21]. Lack of visibility and control in cloud environments can make it challenging for organizations to have full control and visibility into their security posture. Compliance and legal risks may arise from data storage in different geographic locations, subject to different legal jurisdictions and data protection regulations, making compliance with industry-specific regulations challenging [22]. Data loss and leakage can occur due to accidental deletion, hardware failures, misconfigurations in the cloud environment, unauthorized data access, or data sharing with unintended recipients [23,24]. Traditional security measures, such as firewalls and antivirus software, cannot protect cloud environments from sophisticated cyber-attacks [3,25]. By using machine learning, industries can enhance their security measures and reduce the risk of data breaches [26]. The potential of machine learning to boost threat identification and response is one of the main benefits of employing it for cloud security [27]. Traditional security measures such as firewalls and antivirus software are reactive and only respond to known threats. Machine learning, on the other hand, can identify patterns in data that may indicate a threat, even if the threat is not yet known. Based on past data, ML algorithms are generated to find designs that point to security vulnerabilities [28]. A machine learning algorithm can be generated on network traffic data to identify behavioral patterns indicative of a cyber-attack. Once trained, the algorithm may track traffic on the network in real time and notify security staff of any unusual activity patterns. An additional advantage of utilizing machine models for cloud security is its ability to automate certain security tasks [28,29]. Different machine learning algorithms can be recycled to automatically classify and prioritize security alerts, reducing the workload on security personnel [10,30]. It can free security personnel to concentrate on more complex security jobs requiring human expertise. Machine learning can enhance access control and identity management [31]. By examining user behavior patterns, machine learning algorithms can identify out-of-the-ordinary activity that can point to unauthorized access. It can help businesses prevent unauthorized access attempts before they can cause any damage. One of the challenges of employing these models for cloud security is the lack of transparency in how machine learning algorithms make decisions [32,33]. Because machine learning algorithms may be complicated and tricky to read, security employees may find it difficult to comprehend why a certain warning was created [34]. Businesses can use techniques such as explainable AI to address this challenge by making machine learning algorithms more transparent and interpretable. Another challenge is balancing security with usability [35,36]. While security is important, businesses must also ensure that their cloud services are easy to use and do not create unnecessary friction for users. Machine learning can help businesses strike this balance by automating certain security tasks and making security alerts more targeted and actionable [29]. Machine learning has the potential to significantly improve cloud computing security in the industry [37]. By improving threat detection and response, automating security tasks, and enhancing access control and identity management, machine learning can help businesses reduce the risk of data breaches and other security incidents [38]. However, businesses must be aware of the challenges associated with utilizing ML techniques for cloud safety and invest the necessary resources to overcome these challenges [39].

2. Literature Review

Various approaches have been made in the past in which machine learning technology was utilized to enhance cloud computing security. A study examined the ML models to identify and prevent insider threats in the cloud. The study used a dataset of access logs from a cloud platform to sequence a machine learning model to categorize irregular performance indicative of an insider threat. According to the study, the ML model may produce results with a high degree of accuracy in detecting insider threats, demonstrating the potential of machine learning for improving cloud security [40]. Another study explored the use of ML algorithms for improving cloud network security. The study used a dataset of network traffic logs from a cloud platform to develop a machine-learning model to recognize patterns of conduct indicative of a cyber-attack [41,42]. Marwan et al. used SVM and F.C.M. to improve safety in healthcare departments. Their main concern was preventing unauthorized access to medical records, and the suggested approach provides the opportunity to process RGB colorful photos safely using cloud services [42]. Linear regression and SVM boost cyber security by performing the static validation of cloud user action to spot the predefined danger. These security measures are limited in function because they are static [26], designed like the novel security ML model CNN, offering quick and automated solutions to improve security. It offered suggestions for integrating detailed algorithms for securing data throughout all cloud apps instead of solely recognizing sensitive data trends. Mohammad et al. proposed the ML-assisted cloud computing module to enhance safety, and this model exhibited a performance of 95.2%. It also increased the data transmission rates to 96.4% [43]. An attempt was also made to increase the security of the hybrid cloud by using the ML approach, which enhanced C4.5 combined with a duplication algorithm and access control mechanism [43]. The Neuro-fuzzy ML algorithm is used to address cloud cyber security challenges. The ANFIS-dependent variables were largely made to identify anomalies related to cloud privacy [44]. Linear regression (L.R.) and Random Forest (R.F.) algorithms were also used to detect threats and categorize types of attack on multi-cloud environments. The detection and categorization accuracy was 99% and 93.6%, respectively [45].

Novel ML models were designed to cope with security threats automatically [46]. This framework utilized Software Defined Networking (S.D.N.) and Network Function Virtualization (N.F.V.) enablers to reduce various hazards [47]. This AI system combined a monitoring agent with an AI-based response agent that used ML models to analyze network trends and found aberrations that indicated attacks on IoT devices [48]. The modules applied controlled learning, a disseminated information extraction system, and neural networks [49]. ML-based-IDS security solutions were also utilized to prevent threats, network anomalies, and traffic [50]. ML algorithms were also utilized to determine the type of threats. Various supervised, unsupervised, and hybrid threat-determining models were formulated for detecting a threat's type, its purpose (accidental or planned), and motivation (malicious or not) [51,52]. For the purpose of developing security procedures in complicated task handling systems, MLSCS-based multi-server management techniques, RL-Q matrix techniques, V.P.C. setups, and V.G.A.N. were suggested, which proved beneficial [53]. Another machine learning approach based on K.N.N., AdaBoost, and Naïve Bayes was formulated to explore the various security concerns of the cloud environment owing to access, confidentiality, and authenticity. Improved boosting models worked better than K.N.N. concerning accuracy and identification time [54]. The feasibility of two renowned ML strategies, ANN and SVM, were investigated to identify intrusion or unusual behavior in the cloud. SVM showed 91% malware detection accuracy, and ANN exhibited 92% [55]. Thus, many studies have been made on using machine learning for cloud security, such as SVM-based [56], ANN-based studies [57], and MLP-based approaches [58]. The Decision Tree (D.T.) model also proved efficient for detecting anomalies [59]. ML hybrid models have also been studied in the literature due to their superior performances [60,61]. A D.T. and SVM-based hybrid model was proposed to study intrusion detection [62]. The Bayesian-Network-based model was also helpful for managing the cloud. It served to detect attacks [63,64]. The current state of research on employing ML for CC security in the industry demonstrates its potential for improving threat detection and response, automating security tasks, and enhancing access control and identity management.

Machine learning proves to be a promising approach. However, there are still challenges to overcome, such as the need for more transparency in how machine learning algorithms make decisions and the need to balance security with usability. There are several research voids in comparing the effectiveness of ML models for cloud security. This research presents an ML-based strategy for securing cloud computing services. Three machine learning models, SVM, X.G.B., and M.L.P., are proposed in this study. Various hypotheses are proposed in this study, which become the basis for the aims and objectives. The summary of the hypotheses is shown in Figure 1.

Hypothesis 1. 

Machine learning models can be beneficial in determining industrial cloud cyber security, trust, and privacy issues.

In cyber security, machine learning approaches are frequently used to spot vulnerabilities, stop different kinds of assaults, and improve overall security measures. This study addresses cloud security, trust, and privacy issues through a survey from different industrialists, and then ML modules are applied to evaluate this data.

Hypothesis 2. 

Machine learning technology can help identify cyber security and cloud management gaps.

The gaps in cyber security management are important to determine resolutions to the issues of the industrial cloud, and ML can be an effective technique for this purpose. Moreover, there are gaps in utilizing ML technology for the cloud. In this regard, data is obtained from the survey regarding the gaps in cloud security and employment of ML for this purpose, and evaluated by utilizing SVM, XGBoost, and M.L.P. models.

Hypothesis 3. 

SVM, XGBoost, and M.L.P. models can be utilized for the detection of threats and can trigger appropriate actions and mitigate them.

To test this hypothesis, these models are designed, and data regarding security threats obtained from the survey are used as input data to run the models. The performance of these models in determining security threats and triggering appropriate actions is then evaluated using the evaluation matrices.

Hypothesis 4. 

SVM, XGBoost, and M.L.P. modules’ performances vary for cloud computing security.

To test this hypothesis, the performance of these models is compared, and the best-performed method for industrial cloud computing security is determined.

By considering the hypotheses in this study, we set the goals and objectives for this study. The main concern of this study is to formulate a practical strategy for making predictions about the employment of machine learning in the cloud environment. The aims and objectives of the study are as follows.

• Use ML models to sermonize industrial cloud cyber security, trust, and privacy issues.
• Identification of gaps in utilizing the ML approaches for cloud security.
• Detection and mitigation of security threats.
• Triggering appropriate security actions.
• Comparison of the performance of SVM, X.G.B., and ANN models in cloud computing security.

There are numerous studies on the ML approach to cloud security as shown in

Table 1. Literature review summary.

Year	Study	Focus	Key Findings and Limitations
2016	Kaur et al. [32]	Data classification in cloud	Analysis of the security issues at authentication and storage. Development of data classification model. The author does not suggest any framework to solve the security concerns.
2017	Salman et al. [33]	Anomaly detection and classification	Detection of attacks and their classification by LR and RF. 99% detection and 93.6% classification accuracy by RF. It fails to categorize some attacks.
2018	Marwan et al. [34]	Healthcare cloud data security	Prevent unauthorized access to healthcare cloud data. Use of SVM and FCM for image pixel classification to ensure security. It only focuses on image segmentation for security and privacy and does not mention future challenges.
2019	Subramanian et al. [35]	Cloud cyber security	Avoidance of static nature for security verification of the cloud. Use of CNN model for automatic response to threats and saving enterprise data. It does not mention the type of threats, privacy, trust issues, and future challenges for the cyber cloud.
2020	Praveena et al. [36]	Hybrid cloud security	Reduction in security risks to hybrid cloud by enhanced C4.5 algorithm. Determination of the level of security during storing and authorizing data. The author does not discuss threats, trust issues, or future concerns for the hybrid cloud.
2020	Wang et al. [37]	DDOS attack detection	MLP-based model to detect DDOS attacks. Detection based on the feature selection and feedback mechanism to detect errors. The model is not able to find a globally optimized feature. The feedback mechanism can generate a false response.
2020	Chkirbene et al. [27]	Anomaly detection	Classification of scheme to protect the network from unwanted nodes. Reduction of incorrect data issues and differentiation of attacks. The author must discuss trust concerns, industrial cyber issues, and insufficient model comparison.
2021	Haseeb et al. [38]	Health industrial IoT security	Avoidance of uncertainty in data management of the health sector. Data protection by the EDM-ML approach ensures trust between networks. It does not compare the performance of models and does not mention prospects.
2021	Alsharif et al. [39]	IoT security	Use of ML-IDS to take account of traffic defects. The offloading of heavy tasks from the cloud. It does not study industrial cyber cloud concerns or issues regarding the use of the ML approach for the cloud.
2022	Tabassum et al. [33]	QoS security	Neuro-fuzzy approach to studying cloud security, reliability, and efficiency. Discussion of threats, security, and trust issues. No comparison of the model’s performance.
2022	Bangui et al. [40]	Threat detection in Vehicular Ad-hoc Networks (VANET)	Detection and prevention of intrusion in VANET. Use of RF and core set detection for increasing detection efficacy. It does not provide proper solutions to the different types of threats. It lacks performance comparison and trust or privacy factors.

[5]. However, as far as we know, studies have yet to make predictions regarding the employment of machine learning and compare the effectiveness of these three models for industrial cloud security. Additionally, this study offers experimental findings demonstrating the viability of the suggested strategy. It presents a framework that combines feature selection, feature extraction, and classification techniques for detecting anomalous behavior in cloud data. The research methodology involves surveying industry professionals to understand their perceptions of C.C.S. and the potential of ML in enhancing cloud security. The questionnaire includes 15 questions covering various cloud computing security and machine learning aspects. The survey aims to provide insights into the industry’s current security measures, privacy and trust issues, challenges faced, and the potential of machine learning in addressing these challenges. This research concentrates on using ML techniques to improve cloud computing security in the industry. The research explores the current security measures used in the industry and how machine learning can complement these measures to create a more robust and secure cloud computing environment. The paper explores the difficulties in adopting ML for cloud security and suggests solutions.

In conclusion, this paper provides insights into the ability of ML algorithms to boost cloud computing security for industries. This paper highlights the need for continued research and development to create more advanced and efficient security solutions for cloud computing.

3. Methodology

The methodology includes the data collection, evaluation, and architecture for machine learning, which includes model development, platform selection, data transfer, and cloud platform evaluation. The flow sheet for the methodology is given in Figure 2 below.

3.1. Data Collection

Data is gathered from the survey, which consisted of 15 questions covering every aspect of ML in industrial cloud computing security. These survey questions are based on analysis of the literature. The survey is conducted in the major industries of the most populated cities in Pakistan. It depends on various practical factors such as accessibility, data availability, and data collection feasibility within different core industries. It ensures that the necessary permissions and access to collect data from the selected industries will be obtained to gather data regarding the present state of cloud computing security in the industry, the effectiveness of existing security measures, and the potential of machine learning to enhance security. The survey questions are described in

Table 2. Questions for the survey.

Sr No.	Survey Questions
1.	How familiar are you with cloud computing security and machine learning?
2.	Have you or your organization implemented any cloud computing security measures in your operations?
3.	What are the biggest security concerns you have about cloud computing?
4.	How do you think ML can be used to improve cloud computing security?
5.	How confident are you in the effectiveness of current cloud computing security measures?
6.	In your opinion, what are the biggest challenges in implementing effective cloud computing security?
7.	How often do you or your organization conduct security assessments or audits for cloud computing systems?
8.	What role do you think human factors play in cloud computing security?
9.	What measures should cloud service providers take to improve the security of their offerings?
10.	How do you think regulations and compliance requirements affect cloud computing security?
11.	How can organizations ensure their cloud service providers comply with security standards and regulations?
12.	How do you think increasing Internet of Things (IoT) devices affects cloud computing security?
13.	How effectively do you believe machine learning has improved the security of your industry’s cloud computing operations?
14.	How important is security in your industry’s cloud computing operations?
15.	What are your future plans for using machine learning in cloud computing security in your industry?

, and respondent information is given in Figure 3. This questionnaire was shared with 800 experts from different industries via email, and soft reminders were given to them. Almost 400 experts responded to that questionnaire. The respondents’ education is also mentioned to ensure what degree of knowledge they have to answer these critical questions. Most of them are highly educated and are well aware of industrial cloud security issues, so their views are considered reliable. The operational data includes a history of the work completed at each job site. From the operational data, many metrics are derived to reflect a worksite’s current status and past events in a predetermined set of variables. The measurements derived from each hour will serve as an input data vector. A worksite’s finishing time is represented as the number of hours required before completion; hence the output value of an ML model is the number of working hours required before the worksite is finished.

3.2. Experimental Setup

A cloud-hosted version of the Google Colab platform was employed for this endeavor. Windows 10 and a 2.30 GHz Intel Xeon CPU were used for the experiments. Google collaborated with N.V.I.D.I.A. to supply GPUs/CPUs, which were used to speed up calculations. The research was conducted using Python version 3.8, and Google Colab has access to 16 G.B. of RAM for the experiments.

Using a cloud-hosted version of the Google Colab platform for the research has multiple explanations, each leading to the decision to use it. To begin, using a platform hosted in the cloud offers both flexibility and scalability regarding the computing resources that may be utilized. Access to GPUs and CPUs from N.V.I.D.I.A. is made available through Google Colab, and these resources can considerably boost the speed at which calculations are performed and increase the overall quality of the experiments.

In addition, Google Colab features an intuitive interface and integrates nicely with Python, making it a practical tool for carrying out experiments and executing code. In addition, the platform offers a substantial quantity of RAM for the experiments—16 gigabytes—to accommodate more extensive datasets and memory-intensive activities.

3.3. Data Splitting

Data splitting is a common strategy in ML that divides a dataset into two or more subsets, the most common being the training and testing sets. Testing a machine learning model’s performance on data it has never seen before is the primary motivation for data splitting. The training set works to train the data while the results or performance of the model is validated on the testing set. This research split the dataset into two parts. 80% of the data is utilized to run the ML models, while 20% is utilized to analyze the ML models.

3.4. Requirements

The first condition for the cloud service architecture is that all machine learning models, including SVM, XGBoost, and Multilayer perceptron, must be implementable [6,65]. To prevent the need to design architecture on several platforms, all the models must be evaluated on just one platform. The second condition is that the trained algorithm or model must be usable after training for forecasting or prediction.

3.5. Model’s Architecture

This section describes the models used in this research, which are designed for classification, regression, and learning tasks, as the output values obtained from the survey are continuous variables. Several machine learning (ML) models are employed in this study, which are described in this section.

3.5.1. Support Vector Machine (SVM)

The Support Vector Machine (SVM) is the simplest model and requires no hyperparameters. The SVM is a linear classifying method that uses projection to transform the initial dataset into higher-dimensional data. When categorizing subsequent data tuples, the hyperplane is established using support vectors and margins to maximize the hyperplane’s margins in order to generate more precise results [66,67]. Simple linear regression, which has only one input variable, x, and an output variable, y, cannot fulfill the general case where there can be more than one dimension. A multiple linear regression model is utilized, and the equation [67] is given as:

$$y_i={\beta }_o+{\beta }_1{x}_{i,1}+{\beta }_2{x}_{i,2}+\dots {\beta }_k{x}_{i,k}+\epsilon$$

(1)

where, β₁…k = regression coefficient, x_n,1…k = input values, y = output or response, ε = random residual.

3.5.2. Gradient Boosting Model

Gradient boosting models are also used in this paper. The Extreme Gradient Boosting (XGBoost) technique was used to construct gradient-boosting trees. For XGBoost to function effectively, some parameter adjustment is typically necessary. Although the method fundamentally differs from randomization, boosting models build ensembles from numerous single models. Boosting models progressively add fresh single weak models (later known as base learners) to the ensemble [64]. Gradient boosting strategies aim to minimize the predicted value of some loss function L(y, F(x)), such as the squared difference between the true output value y and the output provided by model F(x). K weak base learner’s ϕ(x) is added to form the model F [6,65].

$$F\left(x\right)={\displaystyle \sum }_{k=0}^k{\beta }_k{\varphi }_k\left(x\right)$$

(2)

3.5.3. Artificial Neural Networks

A multilayer perceptron was used to create an artificial neural network. Among the models utilized in this thesis, M.L.P. is the one that needs the greatest fine-tuning [66]. An M.L.P. is a feed-forward system showing that the neurons transmit their outputs forward and do not initiate network cycles [67]. It indicates that a network’s output is a predictable function of its input values. M.L.P. produces a single value from three-dimensional inputs. All input, output, and hidden layers are closely interconnected. Each hidden layer unit uses all three input layer dimensions as inputs, while output layer neurons use all four hidden layer output dimensions as inputs. I and H are the number of input dimensions and hidden layers, respectively. With the activation function in output layer as σ and hidden layer as h, the output is expressed as,

$$y=\sigma \left({\displaystyle \sum }_{J=0}^H{w}_{j^{2}h }\left({\displaystyle \sum }_{i=0}^I{w}_{j{i}^1{x}_i}\right)\right)$$

(3)

3.6. Evaluation Matrices

This research utilized X.G.B., SVM, and M.L.P. machine learning models, several measures from the field of evolution. This research used accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix, which are popular machine learning evaluation metrics frequently used in studies to indicate a model’s performance [6,68]. The following equations determine accuracy, precision, recall values, and F1 score. The model’s effectiveness is studied using a confusion matrix plot. Useful metrics and confusion matrices allow us to assess the total number of correctly and incorrectly labeled classes and precision, recall, accuracy, and F1 score.

$$\mathrm{Accuracy}=\frac{\mathrm{TP} + \mathrm{TN}}{\mathrm{TP} + \mathrm{TN} + \mathrm{FP} + \mathrm{FN}}$$

(4)

$$\mathrm{Precision}=\frac{\mathrm{TP}}{\mathrm{TP} + \mathrm{FP}}$$

(5)

$$\mathrm{Recall}=\frac{\mathrm{TP}}{\mathrm{TP} + \mathrm{FP}}$$

(6)

$$\mathrm{F}1 \mathrm{Score}=\frac{2\times \mathrm{Precision} \times \mathrm{Recall}}{\mathrm{Precision} + \mathrm{Recall}}$$

(7)

4. Results and Discussion

This section describes the results of this work. Machine learning models, including XGB, SVM, and MLP, were applied in this research. A variety of metrics, including accuracy, precision, recall, F1-score, ROC-AUC score, and confusion matrix were utilized in this research. In order to determine how effective the model is, a confusion matrix was drawn up. A total of80% of the dataset is used to train the ML models, while the remaining 20% is used to test the ML models.

4.1. Features Selection

This work aims to establish a practical strategy for making predictions about the employment of ML in the cloud environment. This research created a new dataset to perform the experiments and implement machine learning models for predictions about the employment of ML in the cloud environment. This study utilized this new dataset and applied a feature selection technique to select the most suitable features from the dataset. In this scenario, the feature selection approach (Random Forest Feature importance) is utilized to select the most important features for better prediction. The 11 most convenient features are used in the experiments. The Random Forest Feature importance techniques give a feature importance score for each feature, and according to this score, there are the top 11 features that are most suitable. It is assumed that these features provide valuable information for predicting the target variable or strongly influence the model’s performance.

These features are designed to cover this study’s objectives, which include identifying gaps and uses of ML models in cloud security, trust, and privacy issues. The visualization of each feature score is shown in Figure 4. This study utilized Extreme Gradient Boosting (X.G.B.), Support Vector Machine (SVM), and Multilayer perceptron (M.L.P.).

4.2. ML Analysis

This study uses various ML models and algorithms to evaluate the outcomes. The models are assessed by their accuracy, precision, recall values, and F1 score. The ratio of variables in the model properly predicted to the entire dataset is known as accuracy. The model cannot be judged using just this metric. As a result, it is also important to consider the F1 score, precision, and recall parameters. The performance of the applied models is presented in

Table 3. Machine learning classification results using 11 Features.

Model	Accuracy	Precision	Recall	F1-Score	ROC-AUC
XGB	97.50	97.60	97.60	97.50	1
SVM	97.35	97.30	97.30	97.30	1
MLP	96.20	96.21	96.20	96.20	99

.

4.2.1. XGB Model

The X.G.B. model shows the greatest accuracy score of 97.50%. Additionally, it has the highest precision score of 97.60%, meaning that 97.60% of the model’s positive predictions are true positives. The model demonstrates a recall score of 97.60%, indicating that out of all the actual positive cases, 97.60% are correctly identified by the model. The F1-score is 97.50%, a harmonic mean of precision and recall. The X.G.B. model achieves a perfect ROC-AUC score of one, indicating that the model has excellent predictive power. The ROC-AUC curve is plotted and shown in Figure 5.

The confusion matrix in Figure 6 depicts the X.G.B. model’s performance on a multi-class classification problem. The problem has four classes (or categories) represented by the matrix’s rows and columns. The matrix does not have class labels, but we can infer they are represented by the numbers one, two, three, and four. The diagonal of the matrix displays the number of occurrences successfully categorized for each class. In class one, for example, 33 examples are correctly classified as belonging to class one, whereas one instance is mistakenly classified as belonging to class two. The off-diagonal elements represent the number of misclassifications between the classes, while one occurrence in class three is wrongly identified as belonging to class one. The confusion matrix demonstrates that the X.G.B. model fared quite well on the classification job, with only a few misclassifications.

4.2.2. SVM Model

The SVM model depicts an accuracy score of 97.35%. It displays a precision score of 97.30%, suggesting that 97.30% of the model’s positive predictions are, in fact, true positives. The model shows a recall score of 97.30%, indicating that out of all the actual positive cases, 97.30% are correctly identified by the model. The F1-score is 97.30%, a harmonic mean of precision and recall. The SVM model achieves a ROC-AUC score of one, indicating that the model has excellent predictive power. Figure 7 shows the ROC-AUC curve of the SVM model. The confusion matrix of the SVM model is also plotted and shown in Figure 8.

4.2.3. MLP Model

The M.L.P. model gives an accuracy score of 96.20%. It shows a precision score of 96.21%, demonstrating that out of all the accurate predictions the model made, 96.20% of them are true positives. The model has a recall score of 96.20%, indicating that out of all the actual positive cases, 96.20% are correctly identified by the model. The F1-score is 96.20%, a harmonic mean of precision and recall. The M.L.P. model has a ROC-AUC score of 99%, as indicated in Figure 9, and the confusion matrix, as indicated in Figure 10, demonstrates that the model has excellent predictive power.

Finally, all the models performed well in accuracy, precision, recall, F1-score, and ROC-AUC score. However, the X.G.B. model achieved the highest accuracy, precision, recall, and F1 score among all the models, while SVM and M.L.P. models also performed well.

The study used ML models to describe industrial cloud cyber security, trust, and privacy issues. The objective was to identify gaps in utilizing the ML approach for cloud security, detect and mitigate security threats, and trigger appropriate security actions. In line with these objectives, the study compared the performance of SVM, X.G.B., and M.L.P. models in cloud computing security. The results showed that all the models performed well in accuracy, precision, recall, F1-score, and ROC-AUC score, indicating their potential to address cloud security issues. Specifically, the X.G.B. model achieved the highest accuracy, precision, recall, and F1 score, which suggests that it can be an effective model for detecting and mitigating security threats in the cloud environment. Therefore, the study’s objectives were successfully achieved by demonstrating the potential of ML models in cloud security and identifying the most effective model for this purpose.

5. Conclusions and Suggestions

The employment of machine learning in the cloud environment has been a growing trend in recent years. This research aimed to propose a practical strategy for predicting the employment of ML models in the cloud environment. The study created a new dataset and used a feature selection approach (Random Forest Feature importance) to choose the most important features. The 11 most suitable features were used in the research for the machine learning experiments. The selection of these features proved to help determine ML usage in cloud security, gaps, trust issues, privacy issues, and threat detection, which were the main objectives of this study. The study employed three machine learning models, X.G.B., SVM, and M.L.P., for predictions. The experimental results showed that the X.G.B. model outperformed the other models’ accuracy, precision, recall, F1-score, and ROC-AUC. The proposed approach of using machine learning models for predicting the employment of ML in the cloud environment can be useful for organizations to make informed decisions. The feature selection approach assists in choosing the most crucial features, improving the ML models’ performance. The X.G.B. model showed the highest accuracy and can be used as a reliable prediction model.

Future suggestions could expand this research by adding new features that could enhance the effectiveness of the models. Other ML models, such as Random Forest, Decision Trees, and Naive Bayes, can be utilized in order to compare their performance with the selected models. Furthermore, the dataset can include more instances and various features. It can assist in creating a machine learning prediction model that is more reliable and accurate for use in a cloud setting. User confidence in cloud computing is increased by achieving robust privacy, and algorithm training efficiency can be increased using sophisticated optimization approaches. These techniques will be effective against serious collusion and dependable but observant servers. Using federated learning, investigators and developers can handle various cloud-computing-related problems, such as high communication expenses, confidentiality challenges, statistical variation, and system diversity.