MouadShl/credit-risk-morocco

Readme · MD # 🇲🇦 Credit Card Default Risk Prediction - Moroccan Banking Portfolio > **Production-ready end-to-end ML pipeline** · 30,000 customers · 7 models compared · XGBoost final model · SHAP explainability · Interactive Streamlit dashboard --- ## 🔗 Links 👉 **Launch Live Dashboard** 👉 **View Notebook** 👉 **Executive Report** --- ## 📌 Business Problem Moroccan banks lose **hundreds of millions of MAD annually** to credit card defaults. This project builds a machine learning system that predicts which customers will default **next month** - enabling risk teams to intervene early through: - 💳 Credit limit adjustments - 📋 Structured repayment plan offers - 🚨 Escalation to manual review - 📊 Portfolio-level risk monitoring **Estimated business impact on a 50,000-card portfolio: ~300M MAD saved annually** --- ## 🖥️ Dashboard Preview | Feature | Description | |---|---| | 🎯 Risk Gauge | Live default probability (0–100%) with color-coded tiers | | 🟢🟡🔴 Risk Tiers | Low / Medium / High with branch manager recommendations | | 🔍 Key Risk Drivers | SHAP-based feature contribution bar chart | | 📈 Balance Trend | 6-month statement vs payment chart | | 🏦 Decision Framework | Full 3-tier approval guidelines | | 👤 Sample Profiles | Pre-loaded safe / borderline / high-risk test customers | > Branch managers enter customer data → model predicts default probability in real time --- ## 📊 Results | Model | AUC-ROC | F1 Score | Recall | Precision | |---|---|---|---|---| | Logistic Regression | 0.679 | 0.445 | 0.610 | 0.350 | | K-Nearest Neighbors | 0.643 | 0.190 | 0.121 | 0.435 | | Decision Tree | 0.660 | 0.435 | 0.572 | 0.351 | | Random Forest | 0.684 | 0.433 | 0.518 | 0.371 | | LightGBM | 0.668 | 0.433 | 0.544 | 0.359 | | CatBoost | 0.684 | 0.445 | 0.585 | 0.359 | | **✅ XGBoost (Tuned)** | **0.689** | **0.454** | **0.646** | **0.350** | > XGBoost selected as final model · Tuned via RandomizedSearchCV (40 iterations, 5-fold CV) > Optimal classification threshold: * …