PREDICTING CUSTOMER PENETRATION FOR A BANKING PRODUCT

Abstract

This project aims to develop a predictive model to identify customers most likely to subscribe to a term deposit product, using historical campaign data from a Portuguese bank. Given the class imbalance, SMOTE was applied for data balancing. Multiple models were tested including Logistic Regression, Decision Tree, and Random Forest, on both balanced and unbalanced datasets. Key Findings: • Customers aged 35–60 and those with call durations over 3 minutes are more likely to subscribe. • Success in past campaigns significantly predicts future conversions. • SMOTE-enhanced models outperformed their non-SMOTE counterparts in detecting positive responses. Best Model Selected: Logistic Regression (on SMOTE data) Accuracy: 86.2% | Sensitivity: 95.1% | Kappa: 0.726 Actionable Insight: The model can help the bank reduce irrelevant outreach by focusing only on high- likelihood customers, thus improving customer experience and campaign efficiency.