Train ML models on feature data with walk-forward validation. Models wrap as strategies for backtesting.