An Interpretable Pre-Pruned Decision Tree Framework for Loan Prediction and Customer Segmentation

The growing dependence on machine learning in the financial sector has intensified the need for interpretable and cost-efficient models that can support both prediction and customer segmentation. This study proposes an interpretable framework based on pre-pruned decision trees to address two interrelated tasks: predicting personal loan uptake and segmenting customers into risk-based groups. Unlike traditional rule-based models or black-box ensemble approaches, the pre-pruning strategy constrains tree growth through optimal stopping criteria, thereby improving generalisation, reducing overfitting, and enhancing computational efficiency. Using a real-world banking dataset of 5,000 customers with demographic, financial, and behavioural attributes, the model identifies income, education, and family size as the most influential features through Gini importance. These features guide a hierarchical segmentation process, producing three actionable clusters: low-risk, medium-risk, and high-risk customers. The predictive performance of the pre-pruned decision tree is evaluated against fully grown trees and unsupervised K-means clustering, with results showing that pre-pruning achieves a superior balance between accuracy, F1 score, and interpretability. The findings demonstrate that the proposed framework not only predicts loan uptake effectively but also provides explainable customer segmentation, offering financial institutions a transparent and data-driven tool for targeted marketing and risk management.