Temporal and Multivariate Similarity Clustering of 5G Performance Data

The performance of 5G mobile network cells is highly impacted by their evolving configuration and temporal environmental conditions, such as the number of connected devices or resource utilization. Evaluation of the performance of such a system is a complex task that requires the simultaneous analysis of multiple indicators and inspires the research community to work on zero-touch network service management. In this paper, we present a novel time series clustering method - Temporal and Multivariate Similarity Clustering (TMSC) - that incorporates Dynamic Time Warping with Limited Warping Length and Spectral Clustering, allowing for radio cell grouping based on realization of multiple Key Performance Indicators. We evaluated TMSC against state-of-the-art algorithms at a practical task of identifying cell configuration differences by clustering their performance metrics with a limited set of observations. The proposed algorithm outperformed other methods regarding the Normalized Mutual Information score achieved for more than 95% of the cases studied. We also display the potential for method generalization by evaluating it at the hand gesture recognition task, which yields satisfactory results.