Data-Driven Security: Improving Autonomous Systems through Data Analytics and Cybersecurity

Abstract

This study evaluates the performance and response characteristics of multiple machine learning (ML) models across various cybersecurity threat detection tasks and compared the performance metrics-Accuracy, Precision, Recall, Support Vector Machine (SVM), Random Forest, Neural Network, and K-Nearest Neighbors (KNN) models. Random Forest and SVM demonstrated superior performance, with high accuracy, precision, and recall, and low false positive rates, while KNN lagged slightly. Precision-recall and ROC curves were further analyzed, revealing that Random Forest achieved the highest Area Under Curve (AUC), followed closely by SVM, underscoring their robustness in handling complex data patterns. The data-driven framework outperformed the traditional framework in response time, detection rate, and integration, while the traditional framework exhibited higher user satisfaction. And the response times were analyzed for detecting distinct threat types, including Phishing, Denial of Service (DoS), Malware, and Spoofing. Phishing attacks recorded the lowest response times, while Spoofing and Malware presented higher, more variable times, reflecting their complexity. These results highlight the efficiency of machine learning-based approaches, especially ensemble models, in cybersecurity applications, enhancing detection capabilities and reducing false positives. Our findings provide insights into optimizing model selection and framework deployment to bolster cybersecurity defenses.