REDUCING EDGE NETWORK LATENCY WITH SPATIAL CLUSTERING OF SERVICE NODES: A COMPARATIVE EVALUATION

Edge computing plays a vital role in achieving the low-latency requirements of 5G applications by processing data closer to users. However, sustaining stable latency under heavy or uneven traffic conditions remains a major challenge. This study examines the use of spatial clustering algorithms to organize edge nodes more efficiently and reduce migration delays. A comparative evaluation was carried out on five unsupervised machine-learning algorithms-K-means, agglomerative hierarchical clustering, DBSCAN, mean shift, and Gaussian mixture models-using deployment data from Mosul University's edge network consisting of 50 nodes distributed across a 5×5 km area. In contrast to earlier studies that relied mainly on simulations, the experiments were performed in a live environment, allowing for real-world validation. The findings show that the K-means algorithm achieved the most balanced node distribution with no outliers, while DBSCAN misclassified about 68% of nodes as noise under default settings. During migration tests (from one to five migrations), the latency to the first byte increased by roughly 10 ms per migration, indicating a consistent and predictable relationship between clustering and migration delay. Overall, the results confirm that simple centroid-based clustering can effectively support load balancing in medium-scale edge networks. The study also highlights the need for parameter tuning and statistical validation to ensure reproducible performance. Future work will focus on adaptive clustering methods that respond to dynamic traffic and mobility conditions.