<![CDATA[Customer churn is a significant challenge in the banking industry, which can have a substantial impact on the profitability and long-term sustainability. Customer churn management is typically addressed using binary classification approaches, which often fail to provide the depth needed to understand customer characteristics. This study proposes addressing churn through customer segmentation as an preliminary step before classification, offering a clearer and deeper understanding of each segment’s characteristics. The research combines Self-Organizing Map (SOM) and K-Means clustering to create interpretable segments. The SOM+K-Means model is used for segmentation and visual mapping, which helps identify customer groups at risk of churn and the key features influencing these risks. Cluster labels are then used as features for classification using three machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), and XGBoost (XGB). In the classification phase, the Synthetic Minority Oversampling Technique (SMOTE) and GridSearchCV are applied to address class imbalance and optimize model parameters. XGB outperformed the other models with an accuracy of 85% and an AUC score of 85%. These results highlight that customer segmentation with SOM+K-Means enables more effective churn management strategies, while XGB proves to be a strong model for churn prediction. This research contributes to the application of clustering and machine learning classification techniques in churn analysis within the banking industry, offering a pathway to better customer retention strategies and lower churn rates.]]>
