End-to-end MLB pitching analytics system combining biomechanics, Statcast data, clustering, and machine learning to model pitcher velocity and performance efficiency.
This project simulates MLB R&D workflows to analyze how pitchers generate velocity and how biomechanical + statistical features influence performance.
The goal is not only prediction, but understanding pitcher archetypes and mechanical efficiency.
- Gerrit Cole
- Spencer Strider
- Corbin Burnes
- Justin Verlander
- Zack Wheeler
- Shane McClanahan
Engineered performance features inspired by pitching mechanics:
- Release efficiency
- Movement profiles
- Spin-related metrics
- Velocity differentials
A supervised learning model was built to predict pitch velocity using:
- XGBoost regression
- Feature importance analysis (SHAP)
Unsupervised learning used to identify pitcher archetypes:
- UMAP (dimensionality reduction)
- HDBSCAN (clustering)
Result: grouping pitchers by mechanical similarity.
- Pitch velocity prediction model
- Mechanical efficiency scoring system
- Pitcher archetype clustering
- Scouting-style insights per pitcher
- Automated scouting report (PDF)
Velocity is not only a function of strength, but a combination of:
- mechanics
- release efficiency
- movement patterns
- pitch design
This system helps translate raw Statcast data into actionable scouting intelligence.
- Python
- pandas / numpy
- XGBoost
- SHAP
- UMAP
- HDBSCAN
- pybaseball
- Injury risk modeling (Tommy John risk indicators)
- Pitch tunneling analysis
- Pitch sequencing prediction
- Interactive scouting dashboard (Streamlit)
Built as a sports analytics portfolio project focused on baseball performance analysis and MLB-style data science workflows.