MinneMUDAC2023

Data competition hosted by MinneAnalytics in coordination with the MN Twins

Goal

To forecast game attendance at all regular season MLB games in 2023

Methods

1. Gather historical data from 2012-2022 (around 25K observations)
2. Create several predictive models using training data from 2012-2021 (2020 excluded due to COVID) and with training data from 2022

Data Gathering

Sources

• Retrosheet
  • Game by game records (score, roster, day of the week, day/night game)
  • Hall of Fame players
• Ballparks of baseball
  • Ballpark capacities
  • Cost of construction
• Global sports salaries survey 2019
  • Social media followers (instagram, twitter, facebook, other)
• Swish Analytics
  • Hourly weather reports at the stadium (temperature, percipitation, wind speed)
• Baseball Reference
  • World Series winners and runner ups
• Spotrac
  • Yearly salary averages
• Wikipedia
  • Missing stadium statistics

Variables

There were 40 total variables. The most relevent were the following: avg payroll, season record, number of players on the field who are now in the Hall of Fame, number of total social media followers, region/league, day of the week, year, day/night game

Data gathering methods

Python Webscraper- used for the weather data. The associated Jupyter notebook is in this repo
IA Wayback Machine- used to find salary information that is no longer publicaly accessible
Excel Data from Picture- used to take a picture of a table and quickly transcribe to numbers

Models

The following models were used: XG Boost, PCA Multiple Linear Regression, Regression Tree (JMP), Random Forest