NCAA Modelling: Achieving Tidy Data for March Madness

This analysis details the methodology for parsing, joining, and structuring raw NCAA tournament data into a single, clean “tidy” dataset suitable for training predictive models.

Key Insights

• Data Consistency: Harmonizing team identifiers across multiple historical datasets is critical for preventing leakage and alignment errors.
• Feature Engineering: Calculating rolling team statistics (e.g., offensive efficiency, defensive rating) provides strong predictive signals compared to static historical stats.
• Tidy Principles: Structuring each row as a single matchup with symmetrical team metrics allows standard machine learning algorithms to train effectively.

Check out the complete code, visualizations, and modeling setup in the Kaggle notebook.