Credit Score Classification Using Supervised Machine Learning

Overview

This project involves developing a predictive machine learning system to classify individuals into credit score brackets based on historical credit and financial attributes. The goal is to automate the classification process to support risk assessment and reduce manual efforts for financial institutions.

This is a multiclass classification problem with the target variable being Credit_Score, categorized into three classes:

• 0: Poor
• 1: Standard
• 2: Good

The end-to-end pipeline includes preprocessing, feature engineering, exploratory data analysis (EDA), model training, hyperparameter tuning, performance evaluation, and interpretability via feature importance and ROC curves.

Problem Statement

A global financial organization has collected extensive credit-related data of individuals. The task is to build an intelligent classification system that can predict a person's credit score category (Poor, Standard, or Good) using supervised learning techniques.

Dataset Description

The dataset consists of 100,000 records with 28 features, including demographic data, credit behavior, financial transactions, and account history.

Feature Description

Feature	Description
Age	Age of the customer
Occupation	Job type or profession
Annual_Income	Annual income in USD
Monthly_Inhand_Salary	Monthly salary in hand
Num_Bank_Accounts	Total bank accounts held
Num_Credit_Card	Number of credit cards owned
Interest_Rate	Interest rate on current credit
Num_of_Loan	Number of loans held
Type_of_Loan	Types of loans taken (personal, auto, mortgage, etc.)
Delay_from_due_date	Average number of days late for credit payments
Num_of_Delayed_Payment	Total number of delayed payments
Changed_Credit_Limit	% change in credit card limit
Num_Credit_Inquiries	Number of recent credit checks/inquiries
Credit_Mix	Quality of the credit mix (Good, Standard, Bad)
Outstanding_Debt	Remaining unpaid debt
Credit_Utilization_Ratio	Credit used as a percentage of limit
Credit_History_Age	Duration of credit history in months
Payment_of_Min_Amount	Whether the minimum credit amount was paid
Total_EMI_per_month	Total EMI payments per month
Amount_Invested_monthly	Monthly amount invested
Payment_Behaviour	Credit card payment behavior
Monthly_Balance	Net balance at the end of the month
Credit_Score (Target)	Credit score class (0=Poor, 1=Standard, 2=Good)

Tech Stack

• Language: Python 3.11+
• Development Platform: Google Colab
• Libraries:
  • pandas, numpy, matplotlib, seaborn
  • scikit-learn: model training, evaluation, preprocessing
  • statsmodels: multicollinearity (VIF)
  • GridSearchCV: hyperparameter optimization

Project Workflow

1. Data Preprocessing

• Removal of irrelevant identifiers (e.g., ID, Name, SSN)
• Handling missing values using forward and backward fill
• Type conversions for numeric and temporal fields
• Outlier removal in Age via IQR filtering
• Text parsing of Credit_History_Age to numeric (months)

2. Feature Engineering

• One-hot encoding of categorical features (Occupation, Type_of_Loan, etc.)
• Label encoding of ordinal features (Credit_Mix, Credit_Score)
• Correlation analysis and multicollinearity checks using VIF

3. Exploratory Data Analysis (EDA)

• Distribution plots for income, age, credit score
• Boxplots comparing features across target classes
• Countplots for categorical influence
• Correlation heatmaps for numeric variables

4. Model Development

Four supervised learning models were trained and evaluated:

• Logistic Regression
• Decision Tree Classifier
• Tuned Decision Tree using GridSearchCV
• Random Forest Classifier

All models were trained using train_test_split with an 80-20 ratio and evaluated on unseen data.

5. Evaluation Metrics

• Accuracy, Precision, Recall, F1-Score (macro & weighted)
• Confusion Matrix visualizations
• Feature Importance plots for interpretability
• Multiclass ROC Curves and AUC per class using One-vs-Rest classifiers

Model Performance Summary

Model	Accuracy	Precision (macro)	Recall (macro)	F1-Score (macro)
Logistic Regression	73.5%	0.71	0.71	0.71
Decision Tree	73.2%	0.71	0.71	0.71
Tuned Decision Tree	72.6%	0.70	0.72	0.71
Random Forest	80.2%	0.79	0.79	0.79

Conclusion: The Random Forest Classifier outperformed all other models in terms of both overall accuracy and class-level consistency, making it the most robust model for deployment in this context.

Visual Outputs

• Feature Importance Bar Charts (Logistic & Tree-based)
• Confusion Matrix Heatmaps
• Classification Report Heatmaps
• Multiclass ROC-AUC Curves (per model)

Future Improvements

• Incorporate external credit behavior datasets for feature enrichment
• Experiment with gradient boosting algorithms (XGBoost, LightGBM)
• Apply cross-validation and SMOTE for class imbalance, if needed
• Deploy the model using Flask/FastAPI for production scenarios

Author

This project is part of the Artificial Intelligence Certification Program from Intellipaat and was implemented entirely in Python using Google Colab.

License

This project is for academic and educational purposes. Contact the author for commercial use or collaborations.