Intelligent Customer Acquisition Credit Scorecard Using Explainable AI: LightGBM with ADASYN and SHAP vs. XGBoost with SMOTEENN on the German Credit Dataset

Credit risk assessment is a critical function of financial institutions seeking to identify creditworthy customers during acquisition campaigns. This paper presents the Intelligent Customer Acquisition Credit Scorecard (ICACS), a machine learning framework evaluated on the UCI German Credit Dataset (1,000 records, 20 features).The proposed ICACS framework replaces XGBoost with LightGBM and SMOTEENN with ADASYN (Adaptive Synthetic Sampling), and introduces SHAP (SHapley Additive exPlanations) as a first-class explainability layer. Experimental results demonstrate that ICACS achieves an AUC-ROC of 0.914, outperforming the baseline’s 0.882 by 3.6 percentage points, while also surpassing it in accuracy (78.5% vs. 75.0%), recall (75.4% vs. 69.8%), and F1-score (0.761 vs. 0.715). Crucially, SHAP explainability enables credit analysts to understand and communicate every decision — a capability entirely absent in the baseline. The system is deployed as an interactive Streamlit dashboard supporting real-time individual scoring, portfolio analysis, and policy simulation.