A Novel Density-Based Performance Evaluation Index for Multipath Components Clustering

The machine learning algorithms applied in multipath components (MPCs) clustering should be evaluated by an appropriate performance index. In this paper, a novel performance evaluation index for MPCs clustering is proposed, which is based on the density of clusters instead of cluster distance. The proposed index improves the traditional S_Dbw validity index by taking into account the intra-cluster density obtained according to the Graham scanning method and Green's formula, which can be applied to MPCs with arbitrary distribution characteristics. The proposed index is used to evaluate the performance of different machine learning algorithms, such as K-means and Gaussian mixture model (GMM) algorithms, which shows a more accurate result than other traditional indexes. In addition, the utility of the proposed index is verified by the measured MPCs data involving both delay and angle information. The evaluation result shows that the variational Bayesian GMM (VB-GMM) algorithm outperforms the K-means and expectation maximization GMM (EM-GMM) algorithms in MPCs clustering.