Machine Learning Enhanced Network Intrusion Detection System

Abstract

Contemporary cybersecurity demands have elevated network protection to a fundamental requirement for any computational system. Safeguarding networks from unauthorized infiltration is essential for maintaining seamless operational continuity in advanced network infrastructures. Network protection has emerged as a dominant concern within the information technology domain. Cybercriminals and malicious actors execute countless successful penetration attempts against network systems. An intrusion detection system serves as a cornerstone in network defense, identifying and recognizing irregularities within network security frameworks. IDS effectiveness can be evaluated through its intelligence capacity, operational efficiency, and precise identification of both novel and familiar attack patterns. The maximum gain principleprovides optimal anomaly detection capabilities. This research presents a machine learning architecture utilizing multilayer perceptron (MLP) classification, achieving 99.98% accuracy. The methodology is validated using 10-fold and Jackknife cross-validation techniques. Critical performance indicators, including accuracy, sensitivity, specificity, and Matthew’s correlation coefficient, are analyzed to assess system performance. All evaluation metrics achieved peak performance ratios, demonstrating MLP’s superiority as a classification approach. The proposed model’s accuracy, sensitivity, specificity, and MCC values reached 99.99% when tested on the complete UNSW-NB15 dataset. These findings indicate significant accuracy improvements through various perceptron architectural configurations. Both K-fold and Jackknife methodologies successfully achieved 99.99% accuracy rates.