All projects15 / 43
AI & Data
Race Performance Predictor
Explainable ML for top-3 finishes
At a glance
- Discipline
- AI & Data
- Key features
- 5 delivered
- Screens
- 1 in gallery
PythonXGBoostLightGBMCatBoostOptunaSHAP
Race Performance Predictor
Overview
An explainable model trained on 79,447 race records, combining boosted ensembles, time-aware validation and SHAP explanations.
34 engineered features capture odds, recent form, jockey and trainer performance, track suitability and conditions, with shift and expanding windows to prevent leakage.
Six algorithms were compared, XGBoost was tuned with Optuna, and the strongest boosters were combined into an ensemble explained with SHAP.
Key features
- 0179,447 records across 6,349 races
- 02Time-based splits to prevent future leakage
- 03XGBoost, LightGBM and CatBoost ensemble
- 04Optuna hyperparameter optimisation
- 05Global and per-prediction SHAP explanations
Up next16 / 43
RL Trading Agent
Q-Learning vs Deep Q-Network
