Team Lead: Isaac Wright
Team: Mallet James, Kyle Kroboth, Jeffrey Lunger
Advisor: Andrew Wiesner
Introduction
Since inception, season-long fantasy sports have drawn many casual fans to undertake the competitive task of building and maintaining a virtual roster of professional athletes that compile points through their performance in order to win bragging rights within social circles, test the fan’s own knowledge, and occasionally win money. Within the past few years, lifted restrictions on sports betting around the United States have opened the door for major players to develop new forms of fantasy sports that engage fans on a shorter term basis and keep users coming back daily for fresh excitement, rather than relying on the drag of a season-long league format. With daily fantasy sports, users have the option of choosing new players with assigned salaries every day in order to maximize total points while staying under a budget in global events that offer winners cash rewards. The competition in these events fiercely motivates paying users to constantly improve lineups in order to climb to the top of the leaderboards and win money. While some websites and subscription services offer fantasy advice, few reveal how these recommendations are made, and even fewer build lineups for users. This machine learning model uses baseball data in order to predict the success of individual players, which in-turn maximizes the projected highest scoring lineup under budget for users to submit to daily fantasy games, increasing the odds of winning the cash prizes.
Purpose
In DraftKings daily fantasy sports, the goal is to put together the highest scoring team of players possible when taking into account the cost of selecting each player and the overall budget. The average, everyday fan could select their team based on a variety of factors: basic statistics, pitcher/batter matchups, perceived value, etc. However, any basic user-created lineup is bound to include personal biases. Maybe the user prefers players from their favorite team, former all-stars, younger players, players from large media markets, players they follow on social media, or any other of countless biases that could be incorporated into the fan’s lineup.
The goal of this project is to create a statistical model that will eliminate these biases in lineup-construction, effectively creating the “perfect” lineup based on both current and historical play-by-play data. Our model involves predicting the number of points that each player will score and will utilize an optimization tool that weighs each points projection with that player’s associated cost, and selects the group of players that combine for the maximum number of points while remaining within the budget constraints. We will also suggest the projected highest scoring players, value players based on points and salary, and the players to avoid.
Our goal of eliminating biases in lineup creation links to the overall purpose of the project, which also happens to be the goal of fantasy sports as a whole: to win. We hope to utilize our model to gain an edge on other competitors. For games with cash prizes, our model strives for net positive expected winnings. A successful model would be defined as, “the majority of the games we enter generate an outcome ‘in the money.’”
Tracking the Model
In order to track the success of our model, we will publish our “picks” on each of Friday, Saturday, and Sunday in advance of first-pitch. Following the weekend slate of games, our Monday post will summarize the prediction vs. outcome results. Keep up with each of our updates here.