FraudDetectAI

FraudDetectAI is a machine learning project focused on detecting fraudulent financial transactions. The model achieves an impressive accuracy of 99%, making it a reliable tool for financial institutions and organizations looking to enhance their fraud detection systems.

Table of Contents

• Overview
• Data Cleaning and Preparation
• Model Selection
• Feature Engineering
• Model Performance
• Key Predictive Factors
• How to Use
• Conclusion
• License

Overview

Fraud detection is a critical task in the financial industry. This project employs a RandomForestClassifier to analyze transaction data and identify potential fraud. By cleaning the data, handling missing values, and engineering meaningful features, the model achieves high accuracy and robust performance.

Data Cleaning and Preparation

The dataset used in this project includes features such as Time, Transaction_Type, Amount, Origin_ID, Initial_Origin_Balance, Final_Origin_Balance, Destination_ID, Initial_Destination_Balance, Final_Destination_Balance, Fraud, and Expected_Fraud.

Handling Missing Values

• Missing values in Initial_Destination_Balance and Final_Destination_Balance were filled using the mean of the respective columns.
• The Fraud column's missing values were treated as non-fraudulent transactions.

Outlier Treatment

• The Amount column outliers were handled using the Interquartile Range (IQR) method, ensuring that extreme values do not skew the model's predictions.

Model Selection

A RandomForestClassifier was chosen for its robustness and ability to handle large datasets with high dimensionality. This model was trained on the processed dataset and evaluated using various performance metrics.

Feature Engineering

New features were engineered to enhance the model's predictive power:

• Origin_Amount_Ratio: Ratio of the initial origin balance to the transaction amount.
• Large_Transaction: Binary feature indicating whether a transaction amount exceeds 200,000.
• Hour and Day: Extracted from the Time feature to capture temporal patterns.

Model Performance

The model was evaluated on a test dataset, achieving an accuracy of 99%. Below are the key performance metrics:

• Confusion Matrix

• ROC Curve

• Feature Importance

F1 Score: 99.00%
ROC AUC Score: 99.00%
Accuracy Score: 99.00%

Key Predictive Factors

The key factors that contribute to detecting fraudulent transactions include:

• Transaction amount relative to the origin balance.
• Large transactions that exceed typical thresholds.
• Temporal patterns, including the time of day and day of the week.

Credits

Durvank Gade(https://github.com/DurvankGade) for starting this project.