My Random Forest Was Mostly Learning Time-to-Expiry Noise
3 days ago
- #Feature Importance
- #Machine Learning
- #Random Forest
- Out of Sample Permutation Feature Importance (OOS) is used to optimize features for a Random Forest model.
- OOS involves training the model, permuting feature values in validation data, and evaluating predictive power reduction.
- Gini Importance is criticized for bias towards continuous variables and being computed on training data.
- The model shows an unusually high AUC of 0.7566 for predicting Bitcoin price moves, suggesting potential overfitting.
- The 'seconds_to_settle' feature is identified as overly dominant in the model.
- The model factory has been refactored to use a DSL for easier pipeline configuration and visualization.
- This refactoring aids autonomous agents in discovering and verifying trading strategies.