Unveiling Consumer Patterns Through Predictive Modeling and Clustering on Customer Credit Data

Abstract

This research is motivated by the challenges faced by companies in managing large volumes of customer data to support strategic decision-making. Customer financial data, such as account balances and credit limits, hold great potential for uncovering relevant behavioral patterns but require systematic analysis. This study employs a machine learning approach to cluster customers based on their financial characteristics and to predict payment behavior. The clustering model is implemented using the K-Means algorithm to segment customers into three main groups based on their balances, purchases, and payments. The clustering results are visualized using Principal Component Analysis (PCA) to better understand the distribution among segments. In addition, a predictive model using linear regression is applied to forecast customer payments based on key financial attributes. The research methodology includes collecting credit card customer data, preparing the data through normalization and handling missing values, conducting cluster analysis using K-Means, and evaluating the prediction model using Mean Squared Error (MSE) and R-squared metrics. The dataset used contains financial information from 8,950 entries and 18 attributes of credit card customers. The results show that the K-Means algorithm successfully identified three clusters with distinct financial characteristics. The first cluster consists of customers with moderate balances and high purchase frequency. The second cluster includes customers with low to moderate balances and stable financial activity. The third cluster comprises customers with high balances and credit limits but low purchasing activity. The linear regression model was able to predict customer payments with satisfactory accuracy.