ETD

• additive-models
• machine-learning
• scorecard
• tree-based
• xai

This thesis introduces a framework that combines the predictive power of additive ensembles with the interpretability of scorecards. The method builds on small, low-depth trees, each trained on single features or pairs of features, that are iteratively added to the model. The final predictive system is a transparent scorecard, extracted directly from the trained trees, which preserves human interpretability while maintaining competitive classification performance.