Credit Card Fraud Detection

Project Overview

This project focuses on building and evaluating machine learning models to detect fraudulent credit card transactions. By leveraging a comprehensive dataset of credit card transactions, the goal is to identify patterns indicative of fraud, thereby minimizing financial losses and enhancing security for financial institutions and customers.

Dataset

The dataset used in this project is CreditCard.csv, which contains a large number of credit card transactions.

• Shape: (568,630 rows, 31 columns)
• Memory Usage: 134.49 MB
• Missing Values: None
• Duplicate Rows: None
• Data Types: 29 columns are float64, and 2 columns are int64.

Class Distribution

The dataset exhibits a perfectly balanced class distribution, with an equal number of non-fraudulent and fraudulent transactions.

• Non-Fraud (0): 284,315 (50.00%)
• Fraud (1): 284,315 (50.00%)

Exploratory Data Analysis (EDA)

EDA was performed to understand the characteristics of the dataset, identify important features, and analyze transaction patterns.

Transaction Amount Distribution

The distribution of transaction amounts for both non-fraudulent and fraudulent transactions shows distinct patterns.

• Non-Fraudulent Transactions: Mean amount is 12026.31, median is 11996.90, and standard deviation is 6929.50.
• Fraudulent Transactions: Mean amount is 12057.60, median is 12062.45, and standard deviation is 6909.75.

Feature Importance Analysis

A preliminary Random Forest Classifier was used to identify the most important features for fraud detection. The top 8 most important features are: 'V14', 'V10', 'V12', 'V4', 'V11', 'V17', 'V16', and 'V7'.

Correlation Heatmap of Top Features

A correlation heatmap was generated for the top features to visualize their relationships, including their correlation with the 'Class' variable.

Box Plots of Top Features by Class

Box plots for the top 4 most important features (V14, V10, V12, V4) show their distribution across non-fraudulent and fraudulent classes, highlighting their discriminative power.

Amount vs. Most Important Feature by Class

A scatter plot illustrating the relationship between the transaction Amount and the most important feature ('V14') for both non-fraudulent and fraudulent transactions.

Merchant Category Analysis

[cite_start]Transactions were categorized into 5 bins (Cat_A to Cat_E) based on 'V10' (the second most important feature) to analyze fraud rates across different merchant categories. [cite: 9]

Merchant_Category	Total_Transactions	Fraud_Count	Fraud_Rate
Cat_A	566921	284315	0.5015
Cat_B	1559	0	0.0000
Cat_C	149	0	0.0000
Cat_D	0	0	NaN
Cat_E	1	0	0.0000

• Safest Category: Cat_B, Cat_C, Cat_E (0.0% fraud rate)
• Riskiest Category: Cat_A (50.1% fraud rate)

Transaction Type Analysis

[cite_start]Transactions were categorized by amount into 'Small' (0-100), 'Medium' (100-500), 'Large' (500-2000), and 'Very_Large' (>2000) to understand fraud patterns across different transaction sizes.

Transaction_Type	Total_Transactions	Fraud_Count	Fraud_Rate
Small	1190	594	0.4992
Medium	9599	4719	0.4916
Large	34997	17298	0.4943
Very_Large	522844	261704	0.5005

• Safest Transaction Type: Medium (49.2% fraud rate)
• Riskiest Transaction Type: Very_Large (50.1% fraud rate)

Data Preprocessing

For model training, only the top 8 most important features were selected: 'V14', 'V10', 'V12', 'V4', 'V11', 'V17', 'V16', and 'V7'. Features were scaled using StandardScaler. The dataset was split into training and testing sets (80% train, 20% test) with stratification to maintain class balance.

Machine Learning Models

The following models were trained and evaluated for fraud detection:

• Random Forest Classifier
• Gradient Boosting Classifier
• Decision Tree Classifier
• Gaussian Naive Bayes

Model Performance Metrics

The models were evaluated based on Accuracy, AUC Score, Precision, Recall, and F1-score.

Random Forest

• Accuracy: 0.9783
• AUC Score: 0.9980
• Precision (Fraud): 1.00
• Recall (Fraud): 0.96
• Specificity (Non-Fraud Accuracy): 0.996

Gradient Boosting

• Accuracy: 0.9908
• AUC Score: 0.9994
• Precision (Fraud): 0.99
• Recall (Fraud): 0.99
• Specificity (Non-Fraud Accuracy): 0.993

Decision Tree

• Accuracy: 0.9779
• AUC Score: 0.9965
• Precision (Fraud): 0.98
• Recall (Fraud): 0.98
• Specificity (Non-Fraud Accuracy): 0.977

Naive Bayes

• Accuracy: 0.93
• AUC Score: 0.9850
• Precision (Fraud): 0.99
• Recall (Fraud): 0.88
• Specificity (Non-Fraud Accuracy): 0.989

Summary & Insights

1. Fraud Detection Model Performance:

   • Best Model: Gradient Boosting (AUC: 0.999, Accuracy: 99.1%)
   • All Models:
     • Gradient Boosting: 99.1% accuracy, 0.999 AUC
     • Random Forest: 97.8% accuracy, 0.998 AUC
     • Decision Tree: 97.8% accuracy, 0.996 AUC
     • Naive Bayes: 93.0% accuracy, 0.930 AUC

2. Merchant Category Analysis:

   • Safest Category: Cat_B, Cat_C, Cat_E (0.0% fraud rate)
   • Riskiest Category: Cat_A (50.1% fraud rate)

3. Transaction Type Analysis:

   • Safest Transaction: Medium (49.2% fraud rate)
   • Riskiest Transaction: Very_Large (50.1% fraud rate)