Customer Segmentation and Purchase Behavior Analysis Using Hierarchical Agglomerative Clustering and Machine Learning Techniques

Abstract

The heterogeneity of consumer behavior in digital marketplaces poses a significant challenge for firms seeking to optimize marketing resource allocation and predictive accuracy. This study proposes and validates a hybrid analytical framework that integrates Hierarchical Agglomerative Clustering (HAC) with supervised machine learning techniques to segment customers and predict purchase behavior. Grounded in Machine Learning Theory (MLT) and Computational Learning Theory (COLT), and operationalized through the Cross-Industry Standard Process for Data Mining (CRISP-DM), the research employs a three-stage pipeline: (1) segmentation via HAC using Ward's linkage and Euclidean distance, (2) segment-specific predictive modeling using Choice Model Trees (CMT), Random Forest, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, and (3) comprehensive evaluation using accuracy, precision, recall, F1-score, and ROC-AUC. Using a dataset of 50,000 customer records with demographic, behavioral, and transactional features, the HAC algorithm identified four distinct segments: High-Frequency Power Users, Price-Sensitive Deal Hunters, Low-Engagement Browsers, and High-Value Efficiency Shoppers. The hybrid segment-specific approach achieved a predictive accuracy of 84.33%, substantially outperforming global modeling benchmarks. The Choice Model Trees further provided interpretable decision rules such as the predictive thresholds of five product views and 200 seconds of browsing time enabling actionable managerial insights. The findings demonstrate that addressing customer heterogeneity through hierarchical segmentation prior to predictive modeling yields significant performance uplifts and enhances the interpretability of machine learning outputs for strategic marketing deployment.