Performance Evaluation of Neighbor-Based Learning Models for Network Intrusion Detection System

The rapid advancement of computer networks offers substantial benefits but also introduces growing cybersecurity risks. Cyber-attacks are increasing in both frequency and sophistication, demanding continuous improvements in security systems. Consequently, intrusion detection technologies are being actively researched and optimized to enhance threat detection and prevention capabilities. In particular, Machine Learning (ML) and Artificial Intelligence (AI) have become important tools in enhancing cyber-attack detection capabilities. One of the effective approaches in this field is Neighbor-Based Learning models, a group of ML algorithms that have proven useful in identifying malicious behavior in network traffic. In this study, we apply data preprocessing techniques such as Random Under Sampling to balance the dataset and Robust Scaler to reduce the effect of outliers, thereby improving model performance. Three classification algorithms are implemented, including k-Nearest Neighbors (k-NN), Radius Nearest Classifier (RNC), and Nearest Centroid Classifier (NCC), with the goal of evaluating their effectiveness in detecting cyber-attacks. Experimental results show that after hyperparameter tuning, the RNC model achieves the highest performance with an accuracy of 83.59%, demonstrating strong potential for building efficient and reliable intrusion detection systems. The findings not only contribute to improving detection performance but also suggest new research directions, particularly the integration of Neighbor-Based Learning with Deep Learning or Ensemble techniques to further enhance adaptability in real-time cybersecurity environments.