Enhancing Uplift Analysis through Oversampling Techniques: A Comparative Study on Uplift Models in Cross-Selling

This paper presents a comparative study investigating the impact of oversampling techniques, specifically SMOTE, on the performance of uplift models in cross-selling campaigns with high class imbalance. The study aims to improve the prediction of customers’ likelihood to purchase a new credit account by exploring the application of oversampling techniques. The research covers uplift analysis, introduces commonly used uplift models, and discusses the chosen oversampling technique. Experimental results demonstrate the e↵ectiveness of SMOTE in reducing class imbalance and enhancing uplift model performance. The study concludes by emphasizing the value of oversampling approaches in addressing high class imbalance and providing recommendations for future research and practical implementation in marketing campaigns.

Many thanks for the ING Bank dataset provided by Hon-Prof. Dr. Martin Schmidberger & Dr. Lennart Kraft (Goethe University, Frankfurt Am Main)