Mustafa Ansari, Alexej Schelle

• As e-commerce continues to grow at a rapid pace, the ability to comprehend, segment, and predict customer behavior has become the core of business success. The objective of this project report is to study customer segmentation and predictive modeling using data mining methods on real-life online retail datasets. Exploiting Recency-Frequency-Monetary (RFM), K-Means Clustering, and Predictive Modeling (Logistic Regression, RandomForest, XGBoost, Deep Learning), the investigation explores novel customer segments. It assesses model performance demanding high value from customers. Findings indicate useful implications for segment behavior with the Deep Learning model giving outstanding performance in this case (accuracy up to 87.4% and ROC AUC of 0.932). The RFM-KMeans segmentation approach exposes tactical marketing opportunities in different customer groups, such as Champions, At-Risk Customers, and Big Spenders. This paper proposes a pipeline for scalable and interpretable analyses that combine unsupervised/supervised learning methods to support targeted marketing, retention forecasting, and long-tail customer value maximization in digital retail environments.