Density Clustering Algorithm Based on Shortest Path Maximum Interval

Clustering algorithms represent a crucial research avenue in the realm of data mining. The Density Peak Clustering (DPC) algorithm identifies clusters by pinpointing local density peaks and assigns labels by analyzing the interplay between density and distance. While DPC has made significant strides in identifying local density peaks, it suffers from potential issues such as incorrect center selection and cascading errors in label allocation. To address the shortcomings of the DPC algorithm, this study introduces a novel density clustering approach termed Shortest Path Maximum Interval Density Peak Clustering (SPMI-DPC). This algorithm redundantly selects candidate centers and utilizes the maximum interval of the shortest paths between them to design a new distance metric, \({\delta }_{SPMI}\) , facilitating the construction of a secondary decision graph for more accurate center selection. Additionally, it independently considers the shortest path relationship between each non-center and center to pre- vent cascading assignment errors and ensure accurate label assignment. The algorithm presented in this paper achieves optimal results in 9 out of 10 synthetic and real-world datasets compared to traditional clustering methods and other DPC-based improved algorithms. Among these datasets, of which 8 datasets can achieve 1.00 accuracy, demonstrating excellent clustering performance.