Financial fraud detection is a critical challenge in modern digital financial systems due to the increasing volume, velocity, and complexity of online transactions. This study develops and evaluates a real-time fraud detection framework that integrates distributed cluster computing (Apache Spark and Hadoop) with transaction data mining, combining Interquartile Range (IQR)-based anomaly detection, K-Means behavioural clustering, and four supervised classifiers, Random Forest, Support Vector Machine, Gradient Boosting, and a Neural Network, trained on a 284,807-record transaction dataset. Random Forest, SVM, and Gradient Boosting each achieved 0.98 accuracy, precision, recall, and F1-score, while the Neural Network achieved 0.97 across all metrics, the IQR/K-Means anomaly stage independently achieved 89% precision, 85% recall, and an 87% F1-score. The results demonstrate that combining cluster computing with transaction data mining yields a fraud detection framework that is both highly accurate and scalable for real-time deployment in modern digital financial ecosystems.