Credit Card Fraud Detection

Introduction

Online transactions have become a crucial part of any business over the years. Many of those transactions use credit cards as the main form of payment, and any e-commerce business should be able to protect themselves against fraudulent transactions because such an issue may cost them a lot of money and may have an everlasting impact to the company's reputation.

Many of the fraud detection techniques are rule based and may be simple or more complex, but this approach is susceptible to breach if a clever enough scammer gets to work.

Case Study

In this project we'll use a public credit card transaction dataset that has been anonymized and try a deep learning based approach to create a credit card fraud detection model in order to safeguard an e-commerce business from such dangers.

The challenge posed by this project lies in the fact that most of the transactions will not be fraudulent. This means that the dataset will be heavily imbalanced and maybe a +98% accuracy will not mean that we have a good model at hand. To address this issue we'll rely on resampling, class weighting and properly defined performance metrics to ensure we are training the model correctly.

It's estimated that only 0.1% of business transactions are fraudulent, but even that small proportion can cost a fortune to businesses.

Platform

Google Colab

Programming Language

Python

Topics Covered

• Exploratory Data Analysis (Histograms, Correlation with response, Correlation Matrix)
• Feature Scaling
• Model: Neural Network
• Data Imbalance: Class weighting, SMOTE

Check out the notebook to find out more!