Enhancing security and threat detection through machine learning models

Early threat detection and risk assessment are vital for a proactive guard in network protection, empowering timely interventions to mitigate potential damage and enhance system resilience. This paper examines an AI-driven machine learning (ML) algorithm in enriching internet security using the Canadian Institute for Cybersecurity Intrusion Detection System (CICIDS) dataset of benign and malicious network traffic. It evaluates the efficacy of Random Forest (RF), Support Vector Machine (SVM), and Neural Networks, alongside a conventional rule-based system, based on accuracy, precision, recall, F1-score, and Receiver Operating Characteristic Area Under the Curve (ROC-AUC). Neural networks performed best with 96% accuracy, 95% precision, 96% recall, and a 95.5% F1-score. RF and SVM followed with accuracies of 95% and 92%, respectively, while the rule-based system showed significantly lower performance.