Stochastic Local Search Based Feature Selection Combined with K-means for Clients’ Segmentation in Credit Scoring

Abstract

Segmentation also called clustering is the most important means of data mining. It is an unsupervised learning technique that may be used to split a large dataset into groups. In this work, we propose a new clustering technique that combines the well-known k-means clustering technique with a stochastic local search meta-heuristic. The proposed method is applied to cluster creditworthy customers/companies against non-credit worthy ones in credit scoring. Empirical studies are conducted on five financial datasets. The numerical results are interesting and show the benefits of the proposed technique for banks and clients segmentation.