Credit_Card_Approval_Explainability

Overview

This project focuses on predicting credit card approval using a machine learning model (Multilayer Perceptron, MLP) and interpreting the model's predictions using explainability techniques such as LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations).

The dataset used contains customer features such as income, family size, and mortgage details, with the goal of predicting whether a customer will be approved for a credit card. This project also compares the interpretability of LIME and SHAP in explaining model outcomes.

Features

The dataset contains the following features:

• ID: Customer ID
• Age: Customer Age
• Experience: Work experience (in years)
• Income: Annual income (in thousands)
• Zipcode: Residential area zipcode
• Family: Number of family members
• CCAvg: Average monthly credit card spending
• Education: Education level (1: Bachelor, 2: Master, 3: Advanced Degree)
• Mortgage: Mortgage (in thousands)
• Securities Account: Boolean indicating if the customer has a securities account
• CD Account: Boolean indicating if the customer has a Certificate of Deposit account
• Online: Boolean indicating if the customer uses online banking
• CreditCard: Target column, indicates whether a credit card is approved (1: Yes, 0: No)

Project Tasks

Task 1: Data Loading and Exploratory Data Analysis (EDA)

• Load the dataset and perform basic exploratory data analysis (EDA).
• Visualize and normalize features where appropriate.

Task 2: MLP Model Implementation

• Build an MLP model with no more than 2 hidden layers.
• Perform 5-fold cross-validation to evaluate model performance.
• Report training error and cross-validation error.

Task 3: LIME Explainability

• Select 5 random data points and apply LIME to explain the individual predictions.
• Implement submodular pick and generate LIME explanations for 10% of the training data with up to 5 explanations.
• Use these explanations to predict credit card approval on the entire dataset and calculate the classification error.

Task 4: SHAP Explainability

• For the same 5 data points selected in Task 3, apply SHAP to explain the predictions.

Task 5: Comparison and Observations

• Compare the insights obtained from LIME and SHAP.
• Share key observations on the explainability of the MLP predictions.

Installation

To run this project, you need the following dependencies:

• Python 3.x
• pandas
• numpy
• scikit-learn
• tensorflow
• lime
• shap
• matplotlib
• seaborn

Install the dependencies via pip:

pip install -r requirements.txt

Usage

1. Clone the repository:

   git clone https://github.com/your-username/Credit_Card_Approval_Explainability.git

2. Navigate to the project directory:

   cd Credit_Card_Approval_Explainability

3. Run each py files

   python file_name.py

Results

The results and visualizations of LIME and SHAP will be displayed within the notebook. Additionally, observations on the differences between LIME and SHAP will be discussed in the notebook.