applying hybrid clustering with evaluation by AUC classification metrics

Abstract

Traditional metrics may not adequately assess performance in certain situations, whereas Area Under the Curve (AUC) offers a comprehensive perspective by considering sensitivity and specificity. This method enhances interpretability, addresses limitations, and promotes the development of robust clustering algorithms. Incorporating AUC into unsupervised learning is crucial for refining data analysis strategies. This paper highlights the advantages of using AUC classification metrics as an evaluation tool for clustering algorithms inspired by a recent novel contribution utilizing the AUCC technique. Hybrid clustering models merge the strengths of various clustering approaches to deliver more robust, accurate, and adaptable solutions. Since linkages significantly influence cluster structure and cohesion, selecting the appropriate linkage is essential for accurately discerning patterns in complex datasets. Consequently, we employed single and average linkage methods using Euclidean and Manhattan distance measures. The experiment was conducted on the NSL-KDD dataset for intrusion detection purposes. The results indicated variance in False Alarm Rate (FAR) and Detection Rate (DR) across different NSL-KDD training and testing subsets.