🧠 Credit Risk Modelling using Machine Learning

This project implements an enterprise-grade credit risk assessment system that predicts loan default probability using advanced machine learning techniques, comprehensive statistical analysis, and sophisticated feature engineering methodologies.

The Challenge

• Traditional credit scoring systems lack flexibility and fail to capture complex patterns
• Manual feature engineering is time-consuming and may miss important relationships
• Hyperparameter tuning requires extensive expertise and computational resources
• Model interpretability is crucial for regulatory compliance and business decisions

Our Solution

A state-of-the-art ML pipeline featuring:

• Comprehensive EDA: Statistical tests, distribution analysis, and correlation studies
• Advanced Feature Engineering: WoE (Weight of Evidence) and IV (Information Value) transformation
• Multiple Optimization Strategies: Optuna, GridSearchCV, and RandomizedSearchCV
• Statistical Validation: KS (Kolmogorov-Smirnov) statistic for model performance
• Multiple Encoding Techniques: One-hot, label, target, frequency, and WoE encoding
• Production Deployment: Interactive Streamlit dashboard for real-time predictions

✨ Key Features

📊 Exploratory Data Analysis Statistical Analysis Descriptive statistics Distribution analysis (skewness, kurtosis) Normality tests (Shapiro-Wilk, Anderson-Darling) Chi-square tests for categorical associations Visualization Univariate analysis (histograms, box plots) Bivariate analysis (scatter plots, correlation heatmaps) Multivariate analysis (pair plots, 3D visualizations) Missing Value Analysis Missing data patterns Imputation strategies Impact assessment	🎨 Advanced Feature Engineering WoE & IV Transformation Weight of Evidence (WoE) encoding Information Value (IV) calculation Optimal binning for continuous variables Risk score generation Derived Features Financial ratios (debt-to-income, credit utilization) Aggregated bureau metrics Temporal features Interaction terms Feature Selection IV-based selection (IV > 0.02 threshold) Correlation filtering Variance threshold Recursive Feature Elimination (RFE)
⚡ Hyperparameter Optimization Optuna Bayesian optimization with TPE Multi-objective optimization Pruning for efficiency Visualization of optimization history GridSearchCV Exhaustive parameter search Cross-validation integration Best parameter extraction RandomizedSearchCV Random sampling from distributions Time-efficient exploration Broad search space coverage Comparison & Selection Performance comparison across methods Time vs accuracy trade-offs	🔍 Multiple Encoding Strategies One-Hot Encoding For low-cardinality nominal features Dummy variable creation Label Encoding For ordinal features Maintains order relationships Target Encoding Mean target encoding Smoothing techniques Cross-validation to prevent overfitting Frequency Encoding Category frequency mapping Captures prevalence patterns WoE Encoding Log-odds transformation Strong predictive power for credit risk Business interpretability
📈 Statistical Model Validation KS Statistic Kolmogorov-Smirnov test Measures separation between classes KS > 0.4 indicates good model Performance Metrics Accuracy, Precision, Recall, F1-Score ROC-AUC and Precision-Recall AUC Gini coefficient Confusion matrix analysis Cross-Validation K-Fold validation Stratified K-Fold for imbalanced data Time-series split for temporal data	🚀 Production Deployment Streamlit Application Interactive web interface Real-time predictions User-friendly input forms Visual result presentation Model Persistence Joblib serialization Efficient model loading Version control Scalability Modular code structure Easy integration with APIs Cloud deployment ready

---

📊 Dataset

The project integrates three interconnected datasets for comprehensive credit risk analysis:

Data Sources

Dataset	Description
bureau_data.csv	Credit bureau data with payment history, credit utilization, and delinquency records
customers.csv	Customer demographics including income, employment, and personal details
loans.csv	Loan details with amount, term, interest rate, and default status (target variable)

Key Features by Category

📋 Bureau Data Features

Credit History:

• Credit account age
• Number of active/closed accounts
• Credit mix (mortgage, auto, personal loans)

Payment Behavior:

• Payment history (on-time vs late)
• Number of delinquencies (30, 60, 90+ days)
• Derogatory marks

Credit Utilization:

• Total credit limit
• Total outstanding balance
• Utilization ratio

Inquiries:

• Hard inquiry count (last 6 months)
• Soft inquiry count

👤 Customer Data Features

Demographics:

• Age
• Gender
• Marital status
• Number of dependents

Employment:

• Employment status
• Employment length
• Job title/category
• Income stability

Financial:

• Annual income
• Monthly income
• Income source

Geographic:

• State/City
• ZIP code
• Urban/Rural classification

💰 Loan Data Features

Loan Details:

• Loan amount requested
• Loan term (months)
• Interest rate
• Monthly installment

Loan Characteristics:

• Loan purpose (debt consolidation, home improvement, etc.)
• Loan grade/sub-grade
• Application type

Status:

• Loan status (current, fully paid, charged off)
• Target Variable: Default (0 = No Default, 1 = Default)

Dates:

• Application date
• Issue date
• Last payment date