Credit Card Fraud Detection

This repository contains a project focused on detecting credit card fraud using machine learning techniques. The goal is to develop a reliable model that can distinguish between legitimate and fraudulent transactions, addressing challenges such as data imbalance and ensuring accurate predictions.

Project Overview

Credit card fraud detection is a critical area of focus in the financial industry. This project uses a dataset with various transaction features to identify patterns associated with fraud. The project workflow includes data preprocessing, exploratory data analysis, model training, and evaluation.

Dataset

The dataset used in this project includes the following columns:

• V1 to V28: Unknown features representing principal components obtained from PCA.
• Time: The time elapsed between the transaction and the first transaction in the dataset.
• Amount: The transaction amount.
• Class: The target variable where 0 indicates a legitimate transaction and 1 indicates a fraudulent transaction.

Key Insights:

• The dataset is highly imbalanced, with a majority of transactions being legitimate.
• The Class variable serves as the target for model predictions.
• Fraudulent transactions are typically associated with lower amounts.

Exploratory Data Analysis

During the exploratory data analysis, the following observations were made:

• The dataset's imbalanced nature could lead to overfitting if not addressed.
• Fraudulent transactions generally involve smaller amounts, often below 2500.

Modeling Approach

The project involves the following steps:

1. Data Preprocessing: Handling missing values, feature scaling, and addressing class imbalance using Synthetic Minority Over-sampling Technique (SMOTE).
2. Model Training: Various machine learning models such as Logistic Regression, Decision Trees, and Random Forests are trained on the dataset.
3. Model Evaluation: Models are evaluated based on metrics like accuracy, precision, recall, and F1-score to ensure balanced performance across both classes.

Visualizations

The project includes several visualizations to aid in understanding the data and model performance:

• Count Plots: Show the distribution of legitimate and fraudulent transactions.
• Scatter Plots: Highlight the relationship between transaction amount and fraud likelihood.
• Confusion Matrix: Used to evaluate model predictions.

Getting Started

Prerequisites

Ensure you have the following software installed:

• Python 3.x
• Jupyter Notebook
• Libraries: pandas, numpy, scikit-learn, plotly, imbalanced-learn

Installation

Clone this repository to your local machine:

git clone https://github.com/kinsukh/credit-card-fraud-detection.git

Navigate to the project directory:

cd credit-card-fraud-detection

Install the required dependencies:

pip install -r requirements.txt

Usage

To explore the analysis and run the model, open the Jupyter Notebook:

jupyter notebook final_report_model.ipynb

Execute the cells to perform data analysis, model training, and evaluation.

Conclusion

• The conclusion is that Logistic Regression is the most reliable model for detecting fraudulent transactions in this context, outperforming KNN and Random Forest Classifiers based on key classification metrics.
• Fraudulent transactions are consistently below an amount of 2500, suggesting that fraudsters intentionally target smaller amounts to evade detection.

Results

• The final model shows effective detection of fraudulent transactions, with balanced performance across both legitimate and fraudulent classes. The use of SMOTE helped in mitigating the effects of data imbalance.