FRmicrow/machine-learning

Machine Learning Skill

This skill provides expertise in developing, training, and deploying predictive models for the statFootV3 project.

When to use

Use this skill for tasks related to:

• Designing and implementing football prediction models.
• Training models on historical match and player data.
• Evaluating model performance (Accuracy, Brier score, F1).
• Integrating ML models into the backend via API services.
• Automating data pre-processing for inference.

Hard Rules (CRITICAL)

• Model Choice: Use CatBoost or similar gradient boosting models for structured football data.
• Validation: Ensure proper train/validation/test splits to avoid data leakage.
• Interpretabiltiy: Analyze feature importance to explain model predictions.
• Deployment: Encapsulate models in a service (usually in ml-service/) with a stable API.

Best Practices

1. Feature Engineering: Create meaningful indicators like team form, player ratings (pi-ratings), and xG differentials.
2. Data Cleaning: Handle missing values and outliers carefully.
3. Monitoring: Track model performance over time and re-train periodically with new data.
4. Reproduce: Use seed values for random processes to ensure reproducible results.

Integration

• Works with the data-analyzer skill for data preparation.
• Works with the backend skill for API integration.
• Supports the design skill by providing data for visual storytelling.