Credit Card Fraud Detection: A Hands-On Project

🧑‍🏫 Author: Nhi Yen

💡I write about Machine Learning on Medium || Github || Kaggle || Linkedin. If you found this article interesting, your support by giving me ⭐ will help me spread the knowledge to others.

This project aims to create a model to detect fraudulent transactions in credit card transactions.

Getting Started

Prerequisites

• Python 3.x
• Libraries: pandas, numpy, matplotlib, seaborn, sklearn, scipy, imblearn

Installation

• Clone the repository

git clone https://github.com/username/project.git

• Install required libraries

pip install pandas numpy matplotlib seaborn sklearn scipy imblearn

Usage

Run the following command in the terminal to execute the project:

python main.py

Overview of the Code

1. Exploratory Data Analysis

• Loading data
• Printing random sample of 10 rows to check data loading
• Printing data overview
• Printing numerical summary for Time and Amount columns
• Plotting distribution of Time feature
• Plotting distribution of Amount feature
• Counting number of fraud vs non-fraud transactions and displaying them with their ratio
• Plotting count of fraud vs non-fraud transactions in a bar chart

2. Data Processing

• Plotting heatmap to find any high correlations between variables

3. Modeling

• Drop the 'Class' column to prepare data for splitting
• Get the target variable
• Split data into training, validation and test sets, ensuring the class distribution is maintained
• Initialize the StandardScaler object and fit it to the training data
• Scale the training, validation, and test sets using the scaler
• Undersampling will be utilized to address the issue of imbalanced classes.

Modeling Techniques

1. Logistic Regression

• Run CV with 5 folds (logit)
• Instantiate RandomUnderSampler

2. Naive Bayes

• Fit a Naive Bayes Model

References

• Kaggle Dataset: Credit Card Fraud Detection
• Github Repo - HERE
• Kaggle Project - HERE
• Detail Explanation about the code on MEDIUM

License

This project is licensed under the MIT License.