Study on Empowering Cyber Security by Using Adaptive Machine Learning Methods

Machine Learning (ML) is pivotal in enhancing cybersecurity solutions, surpassing rule-based methods. The complexity of modern malware demands robust detection systems. Traditional signature-based approaches struggle with zero-day and polymorphic threats. Our study presents a versatile ML-based approach adept at identifying active malware and thwarting phishing attempts. We efficiently evaluate web page features to detect malware and phishing attacks by employing ML algorithms in a hierarchical feed-forwarding framework. This approach involves constructing a multilayer model, utilizing ensemble learning techniques in the third layer. Comparative analysis reveals Ensemble Voting (EV) as superior, consistently outperforming Random-Forest (RF), K-Nearest Neighbor (KNN), and Support Vector Machine (SVM) algorithms with an accuracy of 96.16% and a low false-positive rate of 2.6%. Such systems are essential for industrial-level incident detection and security analyst training, emphasizing the indispensable role of ML in constructing dependable cybersecurity infrastructures capable of mitigating evolving cyber threats effectively.