Performance Evaluation of Machine Learning Cluster Metrics for Mobile Network Augmentation

This paper explores the importance of cluster metrics in machine learning (ML), specifically in mobile networks, with a particular emphasis on Device-to-Device (D2D) communication. Cluster metrics are essential in unsupervised learning as they objectively evaluate the performance of clustering methods. Clustering quality is assessed by considering the similarity within clusters and the dissimilarity between clusters. These evaluations are crucial for optimizing algorithm settings, confirming results, and ensuring that clusters correspond to domain expertise. Clustering is essential in mobile networks since it enhances resource allocation, minimizes interference, improves energy efficiency, and facilitates network expansion. This is accomplished by effectively handling intricate communication connections in a well-organized manner. This study investigates popular clustering metrics, such as the Silhouette Coefficient, Calinski Harabasz Index, and Davies Bouldin Index in D2D communication. The usefulness of these metrics is examined by employing several distance measures, such as the physical proximity between devices and data rate metrics and different machine learning clustering approaches like DAIS, FuzzyArt, DBSCAN, GMEANS, MEC, and non-ML approaches like SumRate and Random. Our findings show that traditional network metrics provide a more accurate evaluation because they are related to the primary reason for forming the clusters under the Base Station (BS) in terms of the performance of the clusters. However, the ML cluster metrics must also used to indicate the quality and the clear separation of the devices in a cluster.