Author(s) :

Erick Lumbala Tshipepele, Joel Ilunga Kabuya, Nathanael Mulenda Kasoro.

Volume/Issue :

Abstract :

Consumer attrition remains a major strategic challenge for decision-makers across virtually all sectors. In Kinshasa (Democratic Republic of Congo), within a context of heightened competition and volatile demand, customer loss directly impacts company profitability due to acquisition costs, thereby necessitating a shift from a reactive to a predictive approach in customer relationship management. This article proposes a hybrid methodology for analyzing and modeling consumer attrition, based on behavioral data from supermarket customers in the DRC over a four-year period. The methodology is carried out in two phases: (i) an analytical phase applying Principal Component Analysis (PCA) to enable segmentation and identification of homogeneous profiles in terms of value and churn risk, and (ii) a predictive phase based on supervised classification techniques, specifically decision trees and Support Vector Machines (SVM), to forecast attrition. Models are evaluated using confusion matrices, classification indicators, and ROC–AUC curves derived from supermarket data. The results demonstrate that the proposed approach effectively discriminates at-risk customers, with variables related to purchase intensity and regularity proving particularly decisive. This study highlights the potential of machine learning as a decision-support tool for implementing targeted customer retention strategies in the retail sector in the Democratic Republic of Congo.

No. of Downloads :

0