Implement Dynamic Indexing Strategy for High-Cardinality NoSQL Collections

Design a dynamic indexing mechanism for a MongoDB database that automatically creates and drops indexes based on query frequency and performance metrics. The solution should track query patterns in real-time, use machine learning algorithms to predict optimal index configurations, and minimize write overhead. Implement this as a Node.js middleware that can integrate with existing Mongoose schemas and provides automatic performance tuning.