XGBoost · ML Arena

About XGBoost · ML Arena

XGBoost (eXtreme Gradient Boosting) is a gradient boosting machine learning algorithm, considered the 'king of tabular ML' — the algorithm winning the most Kaggle competitions. On Strategy Arena, XGBoost is trained on hundreds of crypto market features (price, volume, technical indicators, on-chain data) to predict Bitcoin and Ethereum price direction over a 1-24 hour horizon. The algorithm builds a forest of thousands of decision trees, each correcting the errors of the previous one.

How it works

Input features (~100): RSI, MACD, Bollinger, volume, ATR, multi-period EMA, on-chain data (exchange flows, whale moves), hour/day (seasonality). Target: price direction at H+4 (rise > 0.5% = 1, fall > 0.5% = -1, neutral = 0). Training on 6-month rolling window. Temporal cross-validation (walk-forward). Weekly retraining to adapt to new conditions.

Signals used

XGBoost prediction (probability of rise/fall). Feature importance (which indicators carry most weight). Minimum confidence threshold (>60% required to trade). Ensemble of 1000+ decision trees. Weekly retraining.

Risk level

Moderate

Strengths

Top ML algorithm for tabular data (competition-proven). Naturally handles non-linear interactions between features. Resistant to overfitting (built-in regularization). Fast to train and predict. Interpretable feature importance.

Weaknesses

Not designed for time series (no sequential memory). Sensitive to hyperparameter choices. Crypto market changes faster than the retraining window. Risk of data leakage if features aren't properly lagged.

Features

• Category: ML
• Arena: ML Arena
• 24/7 real-time simulation
• Live Binance data