Amelia S

Data Scientist / Quantitative Risk Analyst |

Data Science Nigeria

Jan 2025 - Dec 2025 • 11 mos

Business Context: Partnered with a simulated stakeholder to address the critical tension between predictive accuracy and regulatory interpretability in consumer credit risk management. Model Development & Engineering: Developed 10 predictive models (Logistic Regression, XGBoost, LightGBM, CatBoost, and Stacking Ensembles) using a 30,000-record consumer dataset. Engineered high-impact behavioral features, including recency-weighted delinquency scores and credit utilization ratios, which improved model sensitivity to temporal payment patterns. Performance Optimization: Conducted rigorous hyperparameter tuning using GridSearchCV and stratified k-fold cross-validation to prevent overfitting. Achieved a 3% improvement in AUC-ROC (0.78) over the baseline logistic regression model (0.75). Interpretability & Strategy: Implemented SHAP (SHapley Additive exPlanations) analysis to demystify the "black box" of the ensemble model. Quantified that a two-month payment delay increases default odds by 146% and identified an 80% credit utilization threshold as a critical risk indicator, enabling actionable early-warning strategies.