Train a tree-based machine-learning model (gradient boosting — it combines many simple decision rules into one strong predictor) on features like market category, momentum, liquidity, who created it, and historical base rates, to estimate each market's true resolution probability. You then bet across every market where the model's number beats the price by a set margin. The edge is a data-driven signal that catches nonlinear patterns and feature combinations humans overlook.
Markets on Polymarket where gradient-boosted resolution classifier is the natural play:
