Fraud detection is a critical challenge in financial transactions, requiring advanced machine learning models to distinguish between genuine and fraudulent activities. This project focuses on LSTM-based fraud detection, leveraging historical transaction data to identify suspicious patterns. The model processes multiple attributes, including transaction amount, category, user job type, geolocation, and time-based parameters, to assess fraud risk. In addition to the LSTM model, we conducted single-attribute fraud analysis using various models, evaluating their individual impact on fraud detection. This helped determine the most influential features in predicting fraudulent transactions. The system is integrated into a web-based application built with React and Flask, allowing users to input transaction details and receive a fraud score in real-time. The backend ensures efficient data preprocessing using feature scaling and categorical encoding, aligning new transactions with the trained model’s feature space. Through extensive testing with high-risk and low-risk transaction scenarios, the system demonstrates its ability to detect fraudulent transactions with high accuracy, making it a valuable tool for financial security.