Automated Risk Management Pipeline with Containerized ML Models

Design a Docker-based CI/CD pipeline that automates the deployment and monitoring of machine learning risk prediction models for a trading platform. Create a multi-stage Docker configuration that includes: 1) Data preprocessing with pandas, 2) Model training using scikit-learn, 3) Automated testing with pytest, 4) Kubernetes deployment strategy with auto-scaling based on prediction load. Include comprehensive logging using ELK stack and implement a robust rollback mechanism for model versions. The solution must handle sensitive financial data with encryption at rest and in transit.