MLB Pitcher Strikeout Predictor

Introduction

Strikeouts are one of the most defining statistics in modern baseball. They showcase a pitcher’s dominance, ability to miss bats, and overall effectiveness on the mound. But what if we could predict a pitcher’s strikeout percentage (K%) before the season even begins?

Using historical MLB data from FanGraphs, I developed a machine learning model to estimate each pitcher’s K% for the 2024 season. The model analyzes past performance metrics, plate discipline data, and advanced pitching statistics to generate predictions.

After testing multiple approaches, Ridge Regression proved to be the most accurate model for this task, outperforming other techniques such as Random Forest and XGBoost. Below, I’ll walk through my methodology, the key features that influence strikeout rates, and my final predictions for the 2024 MLB season.

How Strikeouts Are Influenced by Pitching Metrics

Several factors contribute to a pitcher’s ability to generate swings and misses. Based on historical data and advanced sabermetrics, I identified the following key features as the best predictors of K%:

• Fastball Velocity (vFA) – Pitchers who throw harder generally generate more strikeouts.
• Contact% – The percentage of times a batter makes contact when swinging; lower contact rates correlate with higher K%.
• Zone Contact% (Z-Contact%) – The percentage of contact on pitches inside the strike zone.
• Outside Contact% (O-Contact%) – The percentage of contact on pitches outside the strike zone.
• Outside Swing% (O-Swing%) – How often batters swing at pitches outside the strike zone.
• Earned Run Average (ERA) – While not directly related to strikeouts, a lower ERA often indicates better overall pitching quality.
• Games Played (G) – Pitchers who appear in more games generally have more opportunities to record strikeouts.

Building the Strikeout Prediction Model

After compiling the dataset, I tested four different machine learning models to determine which provided the most accurate predictions:

1. Linear Regression (Baseline Model)

• R²: 69.8%
• RMSE: 0.02876
• MAE: 0.02292
• Key Takeaways: Provided a solid baseline but struggled with nonlinear relationships.

2. Random Forest (Nonlinear Approach)

• R²: 65.6%
• RMSE: 0.03071
• MAE: 0.02416
• Key Takeaways: Captured nonlinear trends but had higher error rates than linear models.

3. XGBoost (Gradient Boosting)

• R²: 67.3%
• RMSE: 0.02989
• MAE: 0.02366
• Key Takeaways: Performed better than Random Forest but still fell short of the best model.

4. Ridge Regression (Best Model)

• R²: 69.9%
• RMSE: 0.02867
• MAE: 0.02284
• Key Takeaways: Most accurate model for predicting K%. Effectively handled multicollinearity without overfitting.

Final Predictions for the 2024 MLB Season

Using the final Ridge Regression model, I generated predictions for the 2024 season. The table below highlights five pitchers projected to have the highest K% in 2024:

Conclusion & Future Improvements

This project successfully developed a highly accurate model for predicting MLB pitcher strikeout rates using key performance indicators such as fastball velocity, plate discipline metrics, and earned run average.

With a 69.9% R² and a low MAE of 2.28 percentage points, the Ridge Regression model emerged as the best predictive tool, outperforming nonlinear methods like Random Forest and XGBoost.

Potential Future Enhancements:

• Incorporating new features like pitch spin rate, release point, and pitch usage trends.
• Exploring deep learning models for even better predictions.
• Advanced feature engineering to better capture complex relationships.

Overall, this study provides a data-driven approach to evaluating MLB pitcher strikeout potential, helping analysts, scouts, and fantasy baseball players make better-informed decisions.