Predicting Strikeout Percentage for Relief Pitchers Based on MLB Pitch Data

Luke VanHouten

Baseball Analytics DRP

Winter 2023

Context and Setup

This project is to predict the strikeout percentage of pitchers given information about the pitches they throw as well as the contexts of when they are thrown. The primary source for the data used is a PostrgreSQL database of MLB Statcast pitch data scraped from BaseballSavant. There is a helper R script to build this database included called database.R, which is borrowed from a tutorial created by Bill Petti. The database should be created with the following fields:

user = 'postgres',
password = 'drppassword',
host = 'localhost',
port = 5432

The data should be between 2015 and 2021. I ommitted 2020 data in my original analysis, but this is not necessary. As a result, any predictions made may be different than the ones I made. In order to create this database, simply hit the source button in RStudio to run database.R. Next, for the prediciton script the libraries DBI, RPostgres, RSQLite, tidyverse, and xgboost will need to be installed with the install.packages() function. The baseballr library will need to be installed for numerous helper scripts, as well as the RPostgreSQL and mclust libraries for database.R and clustering.R, respectively.

Running the Model

Simply hit the source button in RStudio for final.R to run the model. It will create numerous dataframes for the different subsets of data used, return the model with the lowest mean absolute error as well as relevant information for it, and create dataframes and a chart detailing feature importance. To plot the distribution of the number of pitch clusters between starters and relievers, run the clusters_plot.R file. Many of the plots created by the model and helper functions can be found in the presentation in the repository. The rest of the helper scripts provide .csv files that have already been created, and thus do not need to be ran.

Article

I wrote a lengthy article about this project on Medium. The article can be found here.

To-Do

• Add 2020 and 2022 years

• Fix position players pitching

• Improve pitch clustering

• Inline comment documentation

Acknowledgements

I am very thankful for statistics PhD student Alex Bank for mentoring me on this project. Without his help I would not have been able to do this project. I am also very greatful for the UW SPA for accepting my application for the project and giving me the opportunity to complete it.