CH.Sai Sushmitha, K.Subash Chandra 03 06 Digital banking adoption has accelerated globally increasing both transaction volumes and the risk of fraudulent activities Traditional fraud detection techniques relying on batch processing and periodic analysis fail to detect fraudulent transactions in real time leading to financial losses and customer distrust This paper proposes a realtime bank transaction fraud detection system that leverages Apache Kafka for streaming transaction data and integrates machine learning algorithms such as Random Forest and Naive Bayes for realtime classification The system architecture ensures scalability fault tolerance and lowlatency processing Experimental evaluation on a largescale bank transaction dataset demonstrates that the proposed system achieves high accuracy precision and recall efficiently detecting fraudulent transactions while minimizing false positives The framework provides a practical solution for modern banking institutions seeking proactive fraud management Copyright (c) Journal of Science Engineering Technology and Management Science. All rights reserved 2026