FRAUD DETECTION IN FINANCIAL TRANSACTIONS

Abstract

This project focuses on developing a machine learning-based fraud detection system using the Paysim dataset to 
identify fraudulent online payment transactions. Traditional fraud detection methods rely on static rules, which 
fail to adapt to evolving fraud tactics. To address this, we propose a dynamic, data-driven approach using the 
Decision Tree Classifier. The system classifies transactions as fraudulent or legitimate based on historical data, 
leveraging features such as transaction amount, location, and user behavior. 
Data preprocessing techniques like handling missing values, feature scaling, and encoding categorical data were 
applied to prepare the dataset for training. The DecisionTreeClassifier model was trained and evaluated using 
metrics such as accuracy, precision, recall, and F1- score. The model achieved high accuracy and was capable 
of detecting fraud with minimal false positives. Additionally, the system integrates real-time fraud alerts and 
performance metrics, allowing administrators to monitor transactions and model performance. 
Future improvements could include integrating advanced machine learning models, such as Random Forests and 
Gradient Boosting, to enhance accuracy. Moreover, handling real-time data and continuous model updates will 
ensure the system remains effective against emerging fraud tactics. This project demonstrates the potential of 
machine learning in creating robust and scalable fraud detection systems.