ADAPTING THE VIRTUAL NOMINAL GROUP TECHNIQUE FOR ENHANCED RISK ASSESSMENT IN CLOUD COMPUTING A MACHINE LEARNING APPROACH FRAMEWORK USING DATA ANALYSIS AND PREDICTIVE MODELLING

Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4802 ADAPTING THE VIRTUAL NOMINAL GROUP TECHNIQUE FOR ENHANCED RISK ASSESSMENT IN CLOUD COMPUTING A MACHINE LEARNING APPROACH FRAMEWORK USING DATA ANALYSIS AND PREDICTIVE MODELLING N. SUJATA KUMARI1 *, SWARNA KUCHIBHOTLA2 Research Scholar¹*, Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh, India, 522502 Associate Professor², Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh, India, 522502 ABSTRACT Given the ongoing evolution of cloud computing platforms and the increasing complexity of cyberattacks, risk assessment is a critical topic. By utilizing algorithmic modeling to forecast risks and altering the traditional Virtual Nominal Group Technique (VNGT), the current study offers an improved method for risk evaluations. The suggested method uses data analysis tools to categorize worry levels, assess possible risks, and offer useful information for proactive risk minimization. The approach enhances cloud security decisions by combining measurable predictive machine learning models with expert-driven subjective assessments. A variety of machine learning algorithms, including supervised and unsupervised methods, are also examined in order to improve the accuracy of risk prediction. Validated on real-world cloud security datasets, the methodology's application shows how well it enhances recognizing risks and remediation tactics. Key Words: Cloud Computing, Risk Assessment, Virtual Nominal Group Technique (VNGT), Predictive Modelling, Machine Learning 1. INTRODUCTION The expandable resources that can be ordered provided by cloud computing have completely changed how companies manipulate, store, and process information [1]. Nevertheless, the greater complex cloud computing systems are, the more susceptible they will be to weaknesses in security including cyber threats [2]. A robust risk analysis strategy that is capable of proactively identifying, assessing, and mitigating any risks is required to provide stored in the cloud infrastructure safety [3]. Stationary standards and subjective evaluation are often the foundation of traditional risk assessment methods, which could prove to be adequate to handle evolving security threats. Conventional cloud computing risk evaluation techniques have a number of issues, such as issues prioritization risks, data overload, and a lack of immediate time alerting of threats [4]. More safety risks could result from the inability of traditional techniques, like systems built around rules or expert based assessments, to identify new cyberthreats. Appropriate ranking of risks is also more difficult for accomplishing when subjective reasoning is used in risk assessment [5]. Driven by information, automated methods that improve turnaround time and danger assessment accuracy are becoming more and more necessary. Using predictive modeling and recognition of patterns with machine learning (ML) presents a viable way to enhance risk identification in cloud computing [6]. ML models are capable of classifying threats based on past privacy occurrences, analyzing enormous amounts of data, and identifying abnormalities. ML can reduce inaccurate results and assist security specialists in detecting risks by combining methods that are both supervised and unsupervised [7,8]. Because of their ability to learn and adjust to new developments in safety reasons, these models are very useful for evaluating cloud risk. This study blends predictive machine learning models coupled with the Virtual Nominal Group Technique (VNGT) to further enhance vulnerability assessment. Using expert opinions and numerical data, VNGT is a methodical approach to decision-making that improves risk prioritizing. Unlike traditional brainstorming methodologies, VNGT mitigates group biases and ensures an increased unbiased risk assessment [9]. This approach can be combined with automatic Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4803 learning driven regulations to provide informed by data integrated expert-validated risk assessment, leading to more secure suggestions. Predictive estimation and data analysis are combined in the approach suggested to evaluate cloud security threats. Computerized learning methods are trained using cybersecurity threat detection databases, which contain information on existing weaknesses, methods of attack, along with severity of risk levels. While regression models generate risk effect scores, classification techniques can forecast the threat's degree of severity. By ranking vulnerabilities according to their prevalence and possible harm, this technique helps organizations adopt more effective methods for risk reduction [10]. Figure 1: Risk estimation approach using Virtual Nominal Group Technique. The figure 1 illustrates an approach for risk estimation based on machine learning methods that incorporates the Virtual Nominal Group Technique (VNGT) towards cybersecurity. 2. RELATED WORKS Cloud computing, with its ability to scale, adapt, and efficiency, has completely changed how businesses manage and organize data. But because cloud systems are dynamic, there are serious security risks, therefore risk assessment is a crucial part of secure cloud management. Conventional methods of assessing risk frequently depend on the subjective and unpredictable expert judgment that is helpful. Researchers have looked into hybrid approaches that combine data-driven procedures with expert knowledge in order to overcome these limitations. These approaches use machine learning and advanced analytics to increase the accuracy of possible detection of risks. The increasing significance underlying machine learning in cybersecurity, especially in anomaly detection and predicting risks, has been emphasized by numerous researchers. While supervised learning techniques like support vector machines and decision trees are commonly employed to identify existing dangers, unsupervised techniques like segmentation and algorithms for detection of anomalies are utilized to find novel and emerging dangers, illustrated in table 1. Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4804 LITERATURE CRITIQUE Prior research has independently examined expert-driven methods such as the Nominal Group Technique (NGT) or prediction models based on machine learning. Nevertheless, little research has been done on combining these strategies to improve cloud risk assessments' contextual relevance and analytical accuracy. Additionally, a lot of models are unable to account for the quick changes in cyberthreats in variable cloud systems. Table 1: Represents the Literature Survey Authors Algorithms Used Pros Dataset Used Other Parameters Smith et.al Random Forest, SVM High Accuracy in Classification Tasks CICIDS201 Precision, Recall, F1 - Score Zhang & Lee K-Means, DBSCAN Effective for Anomaly Detection UNSW-NB15 Clustering Performance, False Positives Kumar et.al Naive Bayes Simple & Interpretable results Custom Cloud Security Logs Computation Time, Accuracy Gupta & Shrama Deep Learning (LSTM, CNN) Improved Pattern Recognition for Evolving Threats CSE-CICIDS2018 Model Training Time, Loss Function Patel at.al Hybrid ML (RF + XGBoost) Better Precision in Risk Evaluation CISA Feature Importance execution Speed Ahmed et.al VNGT + Machine Learning (SVM) Combines Expert Knowledge with Automated Prediction Real World Security Data Subjective vs Objective Assessment Impact Want et. Al Ensemble Learning (Bagging, Boosting) Improved Robustness in risk Scoring KDD99, Cloud Trail Logs ROC-AUC, Model, Interpretability Liu at. Al Decision Tree, Logistic Regression Easy to Implement and Interpret Results AWS Cloud Security Logs Computational Efficiency, Model Simplicity Park & Kim Federated Learning (FL) Enchanting Privacy & Security in Risk Assessment IOT - based Cloud Security Dataset Communication Overhead, Model Convergence Brown et. Al Reinforcement Learning (Qlearning) Adaptive Threat Response Strategy Cloud IDS Logs Training Epochs, Policy Optimization Singh & Verma Gradient boosting (XB Boost) High Performance classification of cloud threats DARPA Intrusion Detection Dataset Hyperparameter Tuning, Feature Selection Rivera et. Al Auto Encoder (Anomaly Detection) Effectively Detects Zero Day Attacks Custom cyber threat intelligence data Reconstruction error, False Alarm Rate Chen et. Al GAN Based Intrusion Learns Adversarial NSL - KDD Training Stability, Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4805 detection Patterns in Cyber Attacks Attack Generation Wilson et. Al Bayesian Networks Probabilistic Reasoning for Threat Analysis Public Cloud Security dataset Bayesian Inference, Prediction Confidence Yadav et. Al Hybrid AI (Deep Learning + Fuzzy Logic) Improves Decision making for risk prioritization Multi Cloud Security dataset Fuzzy Rules, Decision Accuracy 3. METHODOLOGY 3.1 System Model with Problem Statement Rapid changes in cloud computing platforms present opportunities as well as security risks. It is challenging for conventional risk assessment approaches to stay up with emerging hazards due to the growing complexity of malware. The biggest challenge is ensuring the success of a risk assessment technique that combines professional expertise with statistical analysis. The Virtual Nominal Group Technique (VNGT) is not predictive, despite being widely used for expertdriven risk assessment. This work offers a better approach to risk assessment by combining VNGT with modeling that is predictive based on machine learning. The problem statement is to improve cloud safety decisions by employing an advanced approach that efficiently classifies, assesses, and minimizes risks. Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4806 Figure 2: Risk Assessment Framework for Cloud Computing 3.2 Proposed Framework for Risk Assessment in Cloud Computing By incorporating machine learning methods into VNGT, the proposed framework combines qualitative and quantitative in nature risk assessment methods shown in figure 2. It improves the discovery of cloud security risks by fusing expert driven findings with analytical data tools. The framework comprises various components, including risk categorizing, data collection, preprocessing, mitigation strategies, and statistical modeling. Applying unsupervised as well as supervised algorithms to analyze previous threat statistics and generate prospective risk ratings, it assists cloud security organizations in making informed decisions. Figure 3 shows the flowchart. Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4807 Figure 3: Flowchart for Proposed risk assessment framework 3.2.1 VNGT in Cloud Security Risk Assessment Cloud security threat evaluation is greatly enhanced by the Virtualized Nominal Group Technique (VNGT), which employs expert-driven taking decisions and analysis based on data. Conventional risk assessment methods are less successful at predicting and blocking technological advances because systems in the cloud are so dynamic. By using simulated cybersecurity experts, VNGT provides a systematic approach to risk identification, prioritization, and evaluation is. This collaborative approach ensures a range of threats, such as malware attacks, threat actors, security breaches, and mistaken configurations are systematically investigated. Businesses may develop a more comprehensive and adaptable safety structure that is appropriate for the unique challenges posed by cloud computing by fusing risk identification with professional judgment. To further enhance the safety of the cloud assessment of risks, the upgraded VNGT methodology makes use of algorithmic learning (ML) role models, resulting in quantifiable predictive information shows in Figure 4. While researchers independently assess risks based on their practical knowledge, automated learning algorithms both overseen (such as Decision Trees, Random Forests, and Neural Networks) and unattended (Clustering and Anomaly Detection) analyze massive amounts of cloud privacy information to uncover hidden threats and foresee potential vulnerabilities. Using VNGT in conjunction with ML-driven risk analysis can help firms improve secure cloud policies, automate response mechanisms, and more accurately categorize assaults. The efficacy of this method of analysis in mitigating potential hazards and strengthening cybersecurity measures is demonstrated by its reliability on practical applications cloud statistics. Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4808 Figure 4 : VNGT - Based Cloud Security Risk Assessment 3.3 Data Collection The dataset includes network traffic records, ransomware attacks, and penetration attempts from actual cloud security occurrences. The collection of data offers thorough insights into threat trends, facilitating accurate risk assessment and forecasting. Because the data is gathered from many cloud platforms, a wide range of security cases are available for examination. 3.3.1 Data Cleaning To make sure that the dataset utilized for risk analysis is devoid of redundant information and errors, data cleansing is an essential step. Duplicate records are removed, noise is removed, and incorrect values that could impair the model's effectiveness are corrected. To improve information handling efficiency, extraneous data points are also eliminated. The effectiveness of predicting risks is greatly enhanced by making sure that only precise and trustworthy data is used. Cleaned Data = Total Data - Erroneous Data / Total Data * 100 3.3.2 Handling Missing Values Insufficient information can negatively impact the efficiency of models and the precision of predictions. Inadequate information can be handled effectively by a variety of interpolation strategies, including mean/mode replacement, k-nearest neighbors (KNN) attribution, and interpolation methods. The particular kind of imputation approach employed depends on the geographical distribution of missing data. Strong handling techniques are used to ensure that the potential danger prediction model retains its precision as well as data integrity. 3.3.3 Preprocessing The main goal of the proposed work is to increase the accuracy of risk assessment by merging VNGT with predictive machine learning approaches. Computational risk speculation is added to the anecdotal risk assessments provided by experts. To examine threat patterns, a variety of machine learning models are used, such as Random Forest, Decision Trees, Support Vector Machines (SVM), and Neural Networks. Proactive measures for mitigation are made possible by the system's classification of hazards into various severity levels. Data Collection & Integration - The framework collects network traffic information, security logs, and system alarms via actual cloud settings. Unstructured and structured information from a variety of sources, including detection systems for intrusions, network logs, and monitoring cloud tools, are included in this collection. Data Cleaning - Missing values, repeated entries, and mismatched data are addressed via imputation approaches, such as mean/mode restoration for quantitative data and categories range replacement with variable classification. Outlier detection methods, such z-score analysis and Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4809 interquartile range (IQR), can identify and remove anomalous data values that could distort the research. 3.3.4 Supervised learning approaches, like Mean Squared Error (MSE) for regression analyses, cost functions. Technologies for cloud computing are expanding quickly, which offers possibilities for safety as well as concerns. Existing methods of risk evaluation are unable to keep up with fresh weaknesses as cyber attacks become more complex. Developing a successful model for risk assessment that blends analytics driven by data with specialist knowledge is the main problem. Although it is used for expert-driven risk assessment, the Virtual Nominal Group Technique (VNGT) is not predictive. Through the integration of VNGT using automated modeling of future events, this study suggests an improved risk assessment methodology. The challenge is presented as enhancing cloud security choices by using a sophisticated framework that effectively categorizes, evaluates, and reduces risks. 4. MATH FUNCTIONS Table 2 : Math Functions Function Name Mathematical Formula Description Risk Probability Estimation P(Risk / Data) = P(Data / Risk). P(Risk) / P(Data) Predictive modeling is a technique that mostly uses mathematical expressions. The chance of risk occurrence is assessed using Bayesian probability theory. Mean Squared Error MSE=N1∑i=1n(Yi−y^i)2 N - Total number of observations Ai - Actual Value at instance i Âᵢ → Predicted value at instance i Σ → Summation notation (sum over all data points) (Aᵢ - Âᵢ)² → Squared error for each instance Computes the average squared error in machine learning risk models to assess prediction accuracy. Information Gain (Ig = H(Parent) - \Sum \Frac{Child} H(Root) → Entropy of the parent node before splitting H(Child) → Entropy of child nodes after splitting Total Nodes → Number of all instances in the dataset Child Nodes → Number of instances in each subset after the split Σ → Summation over all child nodes Calculator difference between the no of parental and child nodes Euclidean Distance D(X,Y)=∑i=1n(Xi−yi)2 n → Number of dimensions (features) Bᵢ → Value of point B in dimension i Bᵢ → Value of point B’ in dimension i Σ → Summation over all dimensions √( ) → Square root function Aids in the clustering of security threats via unsupervised learning and feature similarity. Journal of Theoretical and Applied Information Technology 15th June 2025. Vol.103. No.11 © Little Lion Scientific ISSN: 1992-8645 www.jatit.org E-ISSN: 1817-3195 4810 Logistic Regression (Risk Score) P(Y)=1+E−(Β0+∑βixi)1 P(Y) → Probability of an event occurring e → Euler’s number (approx. 2.718) C₀ → Intercept term (bias) Cᵢ → Coefficient for feature Dᵢ Dᵢ → Input feature value Use logistic regression to classify security threats and calculate risk scores. Anomaly Detection (Z-Score) Z=X−μ / Σ Z → Standardized value (Z-score) X → Data point value M → Mean of the dataset S → Standard deviation of the dataset Standardizes values to find anomalous activity in cloud security datasets. 5. RESULTS & DISCUSSION To determine how well the applied algorithms performed and how well they classified threats and predicted risks, a total of 18,009 data points were gathered and carefully examined. To assess the cybersecurity risks to cloud networks, a collaborative security architecture was developed. The proposed framework was built in Google Colab using a T4 GPU, which allowed for quicker compute and model training. The system was trained and tested on the UGRansome dataset, which was extracted from Kaggle to look into risk assessment in cloud computing. A range of machine learning techniques, including Random Forest, XGBoost, Support Vector Machine (SVM), and Linear Regression, were used to investigate and forecast security threats. The performance of the XGBoost algorithm was superior to that of the other algorithms. By assessing the results using critical performance metrics such as accuracy, recall, F1score, mean total error (MAE), and mean squared error (MSE), a robust and reliable risk evaluation technique was guaranteed. 5.1 Training & Testing Performances To estimate USD valuations from networked features, a regression job used Linear Regression, Random Forest Regressor, and XGBoost Regressor. A XGBoost Regressor's capacity to reduce predictive mistakes was demonstrated by its lowest MAE and MSE among them. Learning and evaluating the performance of the developed algorithm with an eighty percent training and testing split. The recommended approach successfully learns and distinguishes between dangerous and typical patterns due to its high accuracy and low mistake rates, making it suitable for evaluating cloud safety risks in reality. Table 3: Network Flow Monitoring and IP Address & Threat Classification S.No NetFlow Bytes IP address Threats Port Prediction 1 5 A Bonet 5061 SS 2 8 A Bonet 5061 SS 3 7 A Bonet 5061 SS 4 15 A Bonet 5061 SS 5 9 A Bonet 5061 SS