Fraud Detection Model Using Random Forest
This project presents a fraud detection model developed for financial transactions using a Random Forest Classifier. The dataset contains over 6 million transaction records, with features such as transaction amount, balance changes, and transaction type. The model is designed to identify fraudulent transactions while maintaining high accuracy and recall rates.
Key Features:
Data Cleaning: Handled missing values, outliers, and multi-collinearity.
Model Selection: Random Forest Classifier was chosen for its robustness and ability to handle large datasets and imbalanced classes.
Variable Selection: Important features include transaction amount, old balance, and new balance for both sender and recipient.
Model Performance: Achieved 1.00 accuracy, with a 0.93 precision and 0.73 recall for fraud detection. AUC-ROC score: 0.986.
Visualization: Visuals include confusion matrix, ROC-AUC curve, and feature importance plots.
Fraud Prevention Recommendations: Real-time monitoring, multi-factor authentication, and anomaly detection systems.
Results:
The model effectively detects fraudulent transactions, and recommendations were provided to enhance fraud detection in the company’s infrastructure.