Credit Default Prediction Using Machine Learning Models

Overview:
This project applies machine learning models to predict credit default risks, offering insights to aid financial institutions in making data-driven lending decisions. It focuses on data preprocessing, feature selection, model training, and evaluation.

Project Contents

1. Data Preprocessing and Transformation

• Objective: Prepare a clean dataset by handling missing values, normalizing numerical features, and encoding categorical variables.
• Key Steps:
  • Address missing data using mean imputation for numeric features.
  • Encode categorical variables (e.g., OneHotEncoder for nominal data).
  • Normalize continuous features to standardize scales and prevent any single feature from dominating.

2. Feature Engineering and Selection

• Used Logistic Regression, K-Nearest Neighbors, Decision Trees, Random Forest and XGBoost models to identify the most predictive features, focusing on critical indicators such as loan grade, home ownership status, and financial burden.
• After performing SMOTE to balance the dataset, feature importance was re-evaluated to enhance model interpretability and efficiency.

3. Model Training and Evaluation

• Models trained:
  • Logistic Regression: Baseline model for simple interpretation.
  • K-Nearest Neighbors (KNN): Used for non-linear boundary detection.
  • Decision Trees and Random Forest: To capture complex, non-linear interactions.
  • XGBoost: Primary model due to its robustness in handling imbalanced data.
• Evaluation Metrics: AUC-ROC, F1 Score, Precision, and Recall.

4. Key Findings

• XGBoost demonstrated the highest performance with an AUC of 0.94, balancing recall and precision effectively.
• Feature importance highlighted loan grade and home ownership as critical risk indicators.

5. Figures

• Figure 1: ROC Curve Comparison for Model Performance

• Figure 2: Confusion Matrix for XGBoost Model

6. Key Technologies Used

• Python Libraries: scikit-learn, pandas, NumPy
• Tools: Jupyter Notebook, Colab