Real-time transaction anomaly detection is pivotal in safeguarding financial institutions against fraudulent activities and ensuring the integrity of financial systems. Traditional rule based methods often struggle with scalability and adaptability to evolving fraud patterns, leading to delayed detections and increased false positive rates. This research explores the application of machine learning (ML) techniques in developing predictive models for real time anomaly detection in financial transactions. We evaluate various supervised and unsupervised ML algorithms, including ensemble methods, deep learning models, and hybrid approaches, assessing their effectiveness in identifying suspicious activities with minimal latency. Additionally, we investigate feature engineering strategies and the integration of streaming data processing frameworks to enhance model performance and scalability. Our findings demonstrate that advanced ML models significantly improve detection accuracy and reduce false positives compared to conventional methods, providing a robust framework for real-time financial anomaly detection.