K-means-dist: A Novel Approach for Enhanced Cybersecurity Clustering Using Combined Distance Metrics

Abstract

In the realm of data analysis and cybersecurity, the development of effective clustering techniques plays a pivotal role in understanding complex datasets and identifying anomalies. This paper presents a novel approach based on an extended K-means algorithm that incorporates both correlation and Euclidean distances to improve clustering performance. The method introduces a combined distance function with customizable distance metric weights to capture diverse data relationships. Applied to a cybersecurity dataset (CIC-IDS2018), traditionally used for supervised Intrusion Detection Systems (IDS), we explore unsupervised clustering for IDS, achieving an impressive 92% accuracy. Moreover, our method is adaptable to semi-supervised learning, capturing subtle patterns and relationships in data, making it valuable for various data analysis tasks. This approach has broad applicability for enhancing clustering methodologies across diverse domains.