Design an anomaly detection system for a high-frequency trading platform that processes one million messages per second. Propose an ensemble method combining Isolation Forests and Autoencoders to identify spoofing and layering strategies. Explain how you would handle concept drift in market conditions and minimize false positives in a highly volatile environment without using labeled training data.