Loan Approval/Denial Metrics

Project Description

This project aims to create a an example of a transparent and fair loan approval system and is shown making use of metrics that the SHAP approach relays. This is done using machine learning and explainable AI techniques. The system uses a neural network to predict loan approvals based on applicant features such as age, income, credit score, employment length, and loan amount. To ensure transparency and fairness, SHapley Additive exPlanations (SHAP) are employed to provide clear insights into the decision-making process, highlighting the contribution of each feature to the final decision.

Features

• Data Preprocessing: Cleans and scales the input data to prepare it for model training.
• Model Training: Uses a neural network to learn from historical loan data and make predictions.
• Explainable AI: Utilizes SHAP to provide detailed explanations of model predictions, ensuring transparency.
• Regular Audits: Monitors and audits model performance to detect and mitigate biases.

Example

The SHAP force plot above provides a detailed explanation of the loan approval decision for a specific applicant. Here's what the model output and this plot reveal:

1. Base Value:

   • The base value of 0.6308 represents the average model output if no specific features of the applicant were considered. This is essentially the starting point for the model's decision.

2. Feature Contributions:

   • Age (25): This feature significantly increases the loan approval probability, pushing the value from 0.6308 closer to the final prediction.
   • Credit Score (700): Similarly, a credit score of 700 positively impacts the approval decision, contributing to the increase from the base value.
   • Employment Length (5 years): This feature has a minor negative impact, slightly reducing the likelihood of loan approval.
   • Loan Amount ($15,000): The requested loan amount also contributes negatively, reducing the probability slightly.
   • Income ($50,000): The applicant's income has a minor negative impact, reducing the approval likelihood marginally.

Setup Instructions

Prerequisites

Ensure you have the following installed:

• Python 3.6 or higher
• pip (Python package installer)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/loan-approval-model.git
   cd loan-approval-model

2. Create a virtual environment and activate it:

   python3 -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install the required packages:

   pip install -r requirements.txt

Data Preparation

Ensure you have a dataset in CSV format with the following columns: age, income, credit_score, employment_length, loan_amount, and loan_status.

Place your dataset in the data directory:

mkdir -p data
mv path/to/your/dataset.csv data/loan_data.csv

Usage

Preprocessing Data

Run the data preprocessing script to clean and prepare the data:

python src/data_preprocessing.py

Training the Model

Train the neural network using the prepared data

python src/model_training.py

Explaining Model Decisions

python src/explain_decision.py

Main Application

Run this for the entire application

python app.py

Project Structrue

loan-approval-model/
│
├── data/                     # Directory for the dataset
│   └── loan_data.csv
│
├── model/                    # Directory for the trained model and scaler
│   ├── loan_approval_model.pth
│   └── scaler.pth
│
├── src/                      # Source code directory
│   ├── data_preprocessing.py # Script for data preprocessing
│   ├── model_training.py     # Script for training the model
│   └── explain_decision.py   # Script for explaining model decisions
│
├── pictures/                 # Directory for images
│   └── LoanExample1.png      # Example image of the SHAP force plot
│
├── app.py                    # Main application script
├── requirements.txt          # List of required packages
└── README.md                 # Project README