Customer Segmentation & Sales Strategy System using ML

Abstract

Customer segmentation plays a crucial role in modern retail analytics by enabling data-driven understanding of heterogeneous purchasing behaviors and supporting personalized marketing strategies. Traditional rule-based segmentation approaches often fail to capture complex nonlinear relationships within large-scale transactional data, leading to suboptimal targeting and reduced campaign effectiveness. To address these limitations, a Machine Learning Based Customer Segmentation and Sales Strategy Recommendation System is developed using the Online Retail II UCI dataset covering transactional records with attributes such as product details, quantities, invoice information, customer identifiers, and timestamps. The dataset undergoes extensive preprocessing, including missing value imputation, duplicate removal, and elimination of cancelled transactions, followed by advanced feature engineering to derive behavioral indicators such as Recency, Frequency, Monetary value, basket size, product diversity, and customer lifetime value. Feature scaling is applied prior to model training. Multiple unsupervised clustering techniques, including K-Means Clustering, Agglomerative Clustering, DBSCAN, and Gaussian Mixture Model, are implemented to identify latent customer segments. Model performance is assessed using clustering validation metrics such as silhouette score and Davies–Bouldin index, with K-Means yielding the most stable segmentation results. The optimized model identifies distinct customer groups including Premium, Regular, and Low Value segments, achieving clear behavioral separation. The final system demonstrates 1,577 Premium Customers, 2,512 Regular Customers, and 1,789 Low Value Customers, enabling targeted sales strategy generation and improving customer retention, loyalty targeting, and revenue optimization through datadriven segmentation intelligence.