Predictive Model for Credit Card Defaulting

Using the credit card defaulter dataset, we employed Python techniques and data analysis to discover the patterns behind customer payments and predict whether they would make the payments or not - all within AI4All's Ignite accelerator

Problem Statement

Credit card companies face many risks when customers fail to make their payments. If credit cards are overissued, then there would be a major financial crisis. Hence, the goal of this project is to build a machine learning model that predicts the likelihood of a customer defaulting on their credit card payments.

Key Results

1. Used a dataset consisting of 30,000 different customer data points and cleaned it
2. Based on past payment status and bill statements, it can accurately determine if the customer will pay or not
3. Accuracy rate of 81%

Methodologies

To accomplish this project, we used Google Colab and directly imported the dataset from Kaggle. From there, using Python and data analysis, we cleaned the dataset and removed the unnecessary values that may have caused the model to learn incorrectly. The data was then processed and split into training and testing data, allowing us to detect patterns and develop the model. We used RandomForestClassification so that the model learns from the training data and can classify customers, in order to predict whether they would default or not. We used numpy and sklearn libraries.

Data Sources

https://www.kaggle.com/datasets/gauravtopre/credit-card-defaulter-prediction

Technologies Used

• Google Colab (Python)
• Numpy
• Streamlit

Authors

This project was completed in collaboration with:

Aparnaa Narayanan - aparnaa.narayanan@gmail.com Gwamaka Mwamtobe - mwamtobegwamaka@gmail.com DJ Prichett - djsprichett@gmail.com