Fraud Detection System (ML-Based)

Overview

This project implements a machine learning-based fraud detection system designed to simulate financial transaction monitoring. It combines supervised learning models with anomaly detection to identify suspicious transactions.

Features

• Synthetic dataset generation for transaction data
• Feature engineering based on transaction behavior
• Supervised models:
  • Logistic Regression
  • Random Forest
• Unsupervised anomaly detection using Isolation Forest
• Model evaluation using accuracy, precision, recall, and confusion matrix
• Feature importance analysis for interpretability

Engineered Features

• Amount-to-average ratio: Detects abnormal transaction spikes
• Deviation: Measures difference from typical transaction behavior
• Large transaction flag: Identifies unusually high-value transactions
• Frequency flag: Captures high transaction activity
• Risk score proxy: Combined behavioral risk indicator

Results

• Logistic Regression achieved ~96% accuracy with balanced precision/recall
• Random Forest achieved near-perfect accuracy on synthetic data
• Feature importance highlighted behavioral indicators (ratio, deviation) as key predictors
• Isolation Forest successfully flagged anomalous transactions

Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn

Notes

• Dataset is synthetic and rule-based, used for prototyping and experimentation
• Real-world deployment would require:
  • Noisy, real transaction data
  • Class imbalance handling
  • Model validation on unseen distributions