Loan Payment Status Classification

Overview

This project aims to find the best classifier for predicting loan payment status. We use various machine learning algorithms to analyze a dataset of past loans and determine which borrowers are likely to pay off their loans or default.

Dataset

The dataset (loan_train.csv) contains information about 346 customers whose loans have been either paid off or defaulted. It includes the following fields:

• Loan_status: Whether the loan is paid off or in collection
• Principal: Basic principal loan amount
• Terms: Loan repayment schedule (weekly, biweekly, or monthly)
• Effective_date: When the loan was originated
• Due_date: When the loan is due for repayment
• Age: Age of the applicant
• Education: Education level of the applicant
• Gender: Gender of the applicant

Project Structure

1. Data Loading and Preprocessing
2. Exploratory Data Analysis
3. Feature Engineering
4. Model Training and Evaluation
   • K-Nearest Neighbors (KNN)
   • Decision Tree
   • Support Vector Machine (SVM)
   • Logistic Regression
5. Model Comparison and Selection

Requirements

• Python 3.x
• Libraries: pandas, numpy, matplotlib, seaborn, scikit-learn

You can install the required libraries using:

pip install pandas numpy matplotlib seaborn scikit-learn

Usage

1. Clone the repository:

   git clone https://github.com/yourusername/loan-classification-project.git
   cd loan-classification-project
   

2. Ensure you have the dataset file loan_train.csv in the project directory.

3. Run the Jupyter notebook:

   jupyter notebook find_the_best_classifier.ipynb
   

4. Follow the notebook cells to execute the analysis and model training.

Results

The project compares four different classifiers:

Algorithm	Jaccard	F1-score	LogLoss
KNN	0.7037	0.82	NA
Decision Tree	0.74	0.85	NA
SVM	0.7037	0.82	NA
LogisticRegression	0.76	0.86	0.48

Based on these results, the Logistic Regression classifier performs the best overall, with the highest Jaccard similarity score and F1-score.

Contributing

Feel free to fork this project, submit pull requests, or send suggestions to improve the analysis. All contributions are welcome!