Credit-default-prediction

Credit Scoring Model for Loan Application Assessment

This project involves the development of a predictive credit scoring model to help a financial institution classify unsecured loan applications. By leveraging data analytics, the model aims to minimize credit risk, optimize loan approval rates, and provide a clear understanding of the key factors that drive loan defaults.

Business Problem

A bank needs to refine its loan approval process to balance market competitiveness with risk management. The core challenge is to accurately distinguish between applicants who are likely to repay a loan ("good" customers) and those who are likely to default ("bad" customers).

The key business objectives are:

1. Develop a model that accepts the maximum number of good applicants while correctly identifying at least 85% of bad applicants.
2. Develop a model that accepts at least 70% of good applicants while rejecting the maximum possible number of bad applicants.
3. Identify the most influential variables that determine a customer's repayment behavior to inform future lending strategies.

Dataset

The analysis is based on a historical dataset of past bank customers. The dataset (data.csv) contains customer-level information on financial health and loan characteristics. The target variable is BAD, where:

• 1: The applicant defaulted on the loan.
• 0: The applicant successfully paid off the loan.

Key Predictor Variables Include:

• LOAN: The total amount of the loan requested.
• MORTDUE: The amount due on the applicant's existing mortgage.
• VALUE: The current value of the applicant's property.
• DEBTINC: The applicant's debt-to-income ratio.
• YOJ: Years at the applicant's current job.
• DEROG: Number of major derogatory reports.
• DELINQ: Number of delinquent credit lines.
• CLAGE: Age of the oldest credit line in months.

Analytical Approach

The project follows a structured data analytics workflow to deliver a robust and interpretable solution.

1. Exploratory Data Analysis (EDA):
   • Conducted a thorough investigation of all variables to understand their distributions, identify outliers, and assess data quality.
   • Analyzed relationships between variables and their correlation with the target variable (BAD) using visualizations to rank feature importance.
2. Data Pre-processing:
   • Developed strategies for handling missing values and records.
   • Transformed categorical variables and created new features where appropriate to prepare the dataset for modeling.
3. Modeling:
   • Built and evaluated at least two distinct classification algorithms to predict loan default.
   • Selected appropriate performance measures aligned with the specific business objectives (e.g., recall, precision, ROC-AUC).
   • Incorporated feature selection methods to create parsimonious and effective models.
4. Performance Evaluation and Optimization:
   • Tuned model hyperparameters to optimize performance against the predefined business goals.
   • Assessed the generalization performance of the final recommended models to ensure they are robust and reliable for future predictions.
5. Business Recommendations:
   • Translated the model's findings into actionable business insights.
   • Provided clear recommendations for the bank's lending strategy, highlighting key assumptions and potential limitations of the analytical solution.

Technology Stack

This analysis was conducted using Python and/or SAS. The primary Python libraries used are:

• Pandas for data manipulation and analysis.
• NumPy for numerical operations.
• Matplotlib & Seaborn for data visualization.
• Scikit-learn for building and evaluating machine learning models.

Machine Learning Algorithms Deployed

1. Histogram Gradient Boosting Classifier
2. Random Forest Classifier
3. Support Vector Classifier
4. Logistic Regression
5. k-Nearest Neighbors Classifier
6. Neural Network (Multi Layer Perceptron)
7. Decision Tree Classifier