CatBoost · ML Arena

About CatBoost · ML Arena

CatBoost (Categorical Boosting) is a gradient boosting algorithm developed by Yandex, specifically designed to handle categorical features — day of week, hour of day, market session (Asia/Europe/US), moon phase (!), etc. These categorical features are crucial in crypto because markets have strong seasonal patterns: weekends are more volatile, US market openings impact BTC, Mondays have a bearish bias. CatBoost captures these patterns without manual encoding.

How it works

Natively integrates categorical features: day (Mon-Sun), hour (0-23), session (Asia/EU/US), month, moon phase, calendar events (options expiry, FOMC). Automatic Target Encoding of categories. Ordered Boosting to reduce overfitting (each tree trained on an ordered subset). Same numerical features as other ML models (~100 indicators).

Signals used

CatBoost prediction with integrated categorical features. Seasonality (day, hour, session). Calendar events (options expirations, FOMC meetings, halving). Ordered Boosting (anti-overfitting). Automatic Target Encoding.

Risk level

Moderate

Strengths

Best categorical feature handling (seasonality, sessions). Ordered Boosting effectively reduces overfitting. Captures crypto market temporal patterns (weekend effect, session effect). No need for manual category encoding.

Weaknesses

Slower to train than LightGBM and XGBoost. Crypto seasonal patterns are less stable than traditional markets. The 'moon phase' feature is probably noise (no proven correlation). Marginal advantage vs XGBoost/LightGBM on non-categorical data.

Features

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