Finding the k in K-means Clustering: A Comparative Analysis Approach

摘要

This paper explores the application of inequality indices, a concept successfully applied in comparative software analysis among many application domains, to find the optimal value k for k-means when clustering road traffic data. We demonstrate that traditional methods for identifying the optimal value for k (such as gap statistic and Pham et al.'s method) are unable to produce meaningful values for k when applying them to a real-world dataset for road traffic. On the other hand, a method based on inequality indices shows significant promises in producing much more sensible values for the number k of clusters to be used in k-means clustering for the same road network traffic dataset.