← Back to featured projects
Customer Repurchase Prediction
Machine learning project designed to predict repurchase probability from transactional data. The workflow combines temporal framing, RFM-based features, model optimization and interpretable outputs for customer retention.
PythonLightGBMOptunaDALEXROC-AUC
Best model: LightGBM + Optuna, with ROC-AUC around 0.80 and a clear path to customer prioritisation based on predicted repurchase probability.
What this project shows
- A well-structured supervised learning pipeline with temporal validation logic.
- Feature engineering grounded in RFM and transactional behaviour.
- Model optimization with Optuna instead of a purely baseline workflow.
- Interpretability with DALEX to explain both global and individual predictions.
Assets
One of the strongest business ML projects in the portfolio because it connects evaluation metrics to downstream campaign strategy.