Enhanced supplier clustering using an improved arithmetic optimiser algorithm

Finding the appropriate supplier is becoming increasingly difficult because of the complexity in evaluation criteria. Clustering algorithms and metaheuristic algorithms continue to face challenges while dealing with large data because of their inconsistent convergence, inconsistent balance between exploration and exploitation, and scalability problem. In this paper, we propose an Improved Arithmetic Optimiser Algorithm (IAOA), which integrates population feedback control, directed L´evy search, and cluster-aware soft Lloyd update to improve convergence and avoid local optima trapping. IAOA was tested in two different datasets of textile suppliers: a Moroccan primary case (500 suppliers, 25 Key Performance Indicators) and a Vietnamese transfer case (362 suppliers, 8 KPIs). The Moroccan dataset provides evidence for the existence of five different supplier clusters, where IAOA performs equally to K-means++ and GWOKM. The Vietnamese one does not show consistent discrimination ability. IAOA achieves the best average rank among 11 algorithms in nine benchmark datasets. Ablation analysis shows that the cluster-aware update drives the main performance gain, offering a transparent basis for supplier relationship management.