Robust Real-Time Transaction Anomaly Detection Algorithm

Design a high-performance anomaly detection system for financial transactions using machine learning techniques. The algorithm must process 100,000+ transactions per second with less than 10ms latency, incorporate adaptive learning, and provide a probabilistic risk score. Implement multiple detection strategies including statistical outlier analysis, supervised classification, and unsupervised clustering. Include comprehensive error handling, performance metrics tracking, and a modular architecture that supports easy extensibility for different financial instrument types.