Development of customer segmentation system using supervised and unsupervised machine learning algorithms

Customer segmentation is a crucial strategy in customer relationship management that enables businesses to tailor marketing strategies based on customer characteristics and behaviors. Traditionally, segmentation has been performed using rule-based heuristics such as Recency-Frequency-Monetary (RFM) analysis or demographic grouping. However, these methods often fail to capture complex patterns in large-scale, multidimensional customer data. This study presents the development of a hybrid customer segmentation system combining unsupervised learning (K-Means clustering) and supervised learning algorithms (Random Forest, Logistic Regression) for both discovering customer segments and predicting customer behaviors such as attrition and high-value targeting. To evaluate effectiveness, a real-world customer dataset was used, and the performance of traditional segmentation was compared with the machine learning-based approach. In terms of segmentation quality, Silhouette Score for RFM-based segmentation was 0.34, whereas K-Means clustering achieved a Silhouette Score of 0.62. For predicting attrition using supervised models, the Random Forest classifier achieved Accuracy = 87.3%, Precision = 84.6%, and F1-score = 85.1%, compared to 65.2% accuracy and 58.4% F1-score using rule-based classification. The results demonstrate a significant improvement in both segmentation precision and predictive capability using machine learning approaches over traditional methods. This system enables more data-driven, dynamic, and scalable customer targeting strategies for modern businesses.