Selected solutions from my Kaggle machine learning competitions.
Task: Predict 424 commodity spread targets using time-series modeling.
Approach:
- Time-series ensemble combining LightGBM, XGBoost, and CatBoost
- Lag features and rolling statistics for temporal patterns
- Per-target Ridge regression blending
- 6-fold time-series cross-validation
- Optimized for Sharpe-like rank correlation metric