Prediction of bank customers' purchasing behavior for the optimization of marketing campaigns

This thesis explores the prediction of bank clients' purchasing behavior. The aim was to develop a Random Forest model to optimize marketing campaigns. The model utilized datasets from various digital channels. The data were first merged, cleaned, and enriched with new attributes. Subsequently, a Random Forest classification model was built. The model was optimized using GridSearchCV and validated through five-fold cross-validation. It achieved a ROC-AUC score of 0.8564 on the test set and an overall accuracy of 76%. The F1-score for the purchase class was 0.76. This indicates the model's ability to distinguish between customers with different purchase probabilities. Clients were then segmented into four groups, each receiving a targeted marketing recommendation. The results confirm that analyzing digital interactions can be effectively used to personalize campaigns and increase return on investment.