OPTIMIZING THREAT DETECTION IN CYBER SECURITY USING ML ALGORITHMS

ABSTRACT With the growing sophistication of cyber threats, the demand for timely and effective detection is greater than ever. The current research aims to identify and classify various types of cyber attacks using machine learning algorithms, specifically Naive Bayes, Quadratic Discriminant Analysis (QDA), and Multi-Layer Perceptron (MLP). We also examined attacks like Bots, DDoS attacks, and various Denial of Service variants like GoldenEye, Hulk, Slowhttptest, and Slowloris, and FTP-Patator, PortScan, and SSH-Patator. We chose the most suitable features for each attack type based on feature importance and trained the models accordingly. Our findings revealed that Naive Bayes performed incredibly well consistently, with high accuracy for all types of attacks. It was especially effective in identifying Bots, DDoS attacks, and various DoS attacks, with accuracy rates of over 90% in most cases. Although the MLP classifier displayed excellent performance in some attack types, its performance was highly sensitive to the type of attack. QDA, although helpful, tended to fall behind both Naive Bayes and MLP in accuracy. These results underscore the importance of selecting the most appropriate machine learning model for each type of threat. Naive Bayes proved to be a especially robust tool for real-time identification of threats and thus a valuable addition to the enhancement of cybersecurity strategies. This research underscores the importance of customized approaches in machine learning for cybersecurity and provides practical insights into the optimization of threat detection and response strategies. Keywords: Machine Learning, Threat Detection, Naive Bayes, Quadratic Discriminant Analysis, Multi-Layer Perceptron, Bot Detection, DDoS Attacks, DoS Attacks, Feature Importance, Attack Classification.