ML Ensemble BTC 5-Min Strategy
Uses an ensemble of LGBM, XGBoost, and LSTM models to predict BTC Up/Down direction on 5-minute Polymarket markets, with automated training via GitHub Actions.
This bot trades the BTC Up/Down 5-minute binary option markets on Polymarket. It employs a three-model ensemble: LightGBM (gradient boosting), XGBoost, and an LSTM neural network. Features likely include recent price action, order book imbalances, volume, and technical indicators. The ensemble outputs a probability of 'UP' vs 'DOWN', and the bot takes a position when confidence exceeds a threshold. Training is fully automated via GitHub Actions scheduled jobs, retraining the models on fresh data as new market outcomes resolve. The model's edge comes from capturing non-linear patterns and regime changes that simple logistic models miss. The repo includes a live dashboard on GitHub Pages for monitoring performance, win rate, and P&L. This strategy is computationally lightweight and can run serverlessly.