Predictive Modelling of Customer Churn in a Nigerian Commercial Bank Using Decision Trees

Customer churn imposes significant revenue losses on Nigerian commercial banks operating in an increasingly competitive retail banking environment, and predictive modelling offers an evidence-based tool for proactive retention management. This study developed a decision tree model to predict customer churn in a Tier 2 commercial bank operating in Lagos and Ibadan, South West Nigeria. De-identified customer transaction records for 12,000 retail accounts were provided by the bank under a data sharing agreement. Features included average monthly balance, transaction frequency, product portfolio breadth, customer tenure, loan repayment history, and customer service interaction counts. Data preprocessing addressed class imbalance using SMOTE oversampling, achieving a balanced training set. The CART decision tree model achieved an AUC of 0.83 and an F1-score of 0.79 for the churn class. The most important predictors were a decline in transaction frequency over three consecutive months, zero cross-product holding, and absence of digital banking engagement in the preceding 90 days. The model correctly identified 76.4 percent of churned customers in the validation set. Misclassification rates were highest for customers who maintained savings accounts while withdrawing from other products. Results suggest that cross-selling and digital engagement monitoring are among the most actionable retention levers for the bank. Recommendations include implementing a real-time churn score dashboard for relationship managers, designing targeted retention interventions for high-risk segments, and expanding the feature set to include customer satisfaction survey data.

Keywords: customer churn prediction, decision tree, commercial bank, Lagos, SMOTE