Background
As someone who grew up playing football, watching football on Saturdays and Sundays, and is a regular fantasy football participant, I’ve been only experienced football as a player and fan. I’ve never gotten the opportunity to experience what it is like to experience football from the business perspective. However, over this past summer, I got this opportunity as I got work on a project in collaboration with a business analytics team of an NFL organization. The work that I conducted mainly entailed gaining insights and providing feedback into a recently rolled out rewards program for the organization’s season ticket holders. To see how the program stands in its initially years, the goal of the project was to gain insights into spending and usage among the season ticket members.
Project Details
The data that I worked was provided by the business analytics team. The datasets that were provided included information about the season ticket holders which mainly focused on attendance and transactional data over the past few seasons. Little data cleaning or manipulation was needed but imputing missing data values and extracting JSON formatted data. These were resolved by KNNImputer to provide better estimates of the missing data and regular expressions to extract the necessary data from the JSON format by recognizing patterns in the format.
After cleaning the data, the focus of my work was creating models to predict season long spending, use of the rewards program, and vending type preference. The model that I choose to implement for these was the random forest classifier model. The way that the random forest classifier works is that it gets trained on a portion of the data and creates hundreds of decision trees that create different splits on the data being trained with the goal of isolating classes into their own cluster or group. Each model had a slightly different approach. The approach taken to modeling season long spending was to predict which percentile bucket that a season ticket member will end up in. The approach taken in modeling the use of the rewards program was a binary decision of whether a season ticket member in attendance uses the rewards at least once during the duration of the game. The approach taken in modeling the vending preference was looking at what a season ticket member tends to purchase to predict their most used vending preference.
Given the scope of the project, I was able to provide insights including at what point relative to the starting point of the game are season ticket members arriving to the stadium, at what point are they making transactions, what types of food are more prevalent in transactions, and whether there are differences in the transactions being made by vending type. Furthermore, modeling gave insights into the factors that contribute to different levels of spending, vending preferences, and program usage. These insights are useful in that it allows for an understanding of preferences and tendencies of the organizations season ticket holders, which in turn will allow the organization to better understand the usage of the program.
#SportsInnovation #SportsAnalytics #Indy4Sports
Leave a Reply