An Improved Density Peak Clustering with Flexible Manifold Distance and Natural Nearest Neighbors for Network Intrusion Detection

Recently, the field of density peak clustering (DPC) has garnered attention for its ability to intuitively determine the number of classes, identify arbitrarily shaped classes, and automatically detect and exclude anomalies. However, DPC faces challenges in considering only global distribution, resulting in difficulties with group density, and its point allocation strategy may lead to a domino phenomenon. In order to give DPC a broader scope to showcase its talents, this paper suggests a Density Peak Clustering algorithm based on Manifold Distance and Natural Nearest Neighbors (abbreviated as DPC-MDNN), which constructs a nearest neighbor relationship based on manifold distance and introduces representative points using local density to segment the distribution. It adopts an assignment strategy based on representatives and candidates, reducing the domino effect through micro cluster merging. Extensive comparisons with five competing methods on both artificial and real datasets demonstrate that DPC-MDNN can identify clustering centers more accurately and achieve better clustering results. Furthermore, application experiments on two sub-datasets have confirmed that DPC-MDNN can improve the accuracy of network intrusion detection and has high practicality.