Out of Sample Permutation Feature Importance For Random Forest’s Feature Optimization
Summary
The post explores using Out-of-Sample Permutation Feature Importance to optimize a Random Forest after initial parameter tuning. It finds that a time-based feature (seconds_to_settle) dominates and may indicate lookahead bias, leading to aggressive feature pruning and refactoring. The author also introduces a DSL-powered model factory and discusses implications for autonomous trading strategies.