Performance Analysis of Mobile Wireless Networks Through Crowdsourcing Data Clustering

Crowdsourcing data has emerged as a valuable resource for Mobile Network Operators (MNOs), offering user-centric insights about network performance. This paper introduces a methodology for identifying network performance patterns in crowdsourced data using a combination of clustering algorithms and Clustering Validity Indexes (CVIs). The methodology integrates well-established clustering algorithms - K-means, Spectral Clustering (SC), Gaussian Mixture Model (GMM), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) - with the Euclidean Distance-Optimized (EDO) transformation, which improves clustering results by addressing non-normal distributions within the performance variables. The proposed approach applies multiple CVIs to evaluate the quality of clustering partitions, ensuring a comprehensive evaluation of the crowdsourcing dataset. Then, the optimal clustering solution is selected through a voting mechanism, with the SC emerging as the selected one. The analysis revealed two distinct performance patterns resulting from the use of different Radio Access Technologies (RATs), with the EDO transformation proving to be more effective, yielding higher-quality cluster partitions compared to conventional z-standardisation. The proposed methodology represents a structured step toward augmenting the application of crowdsourcing data for network performance analysis and optimisation.