In Daily Fantasy Sports (DFS) contests, contestants construct a virtual lineup of players that score points based on their real-world performances. Unlike in season-long Fantasy Sports contests,in DFS contestants submit a new lineup for each set of games. DFS contests are held for several professional sports leagues, including the National Football League (NFL), National Basketball League (NBA), and National Hockey League (NHL). The leading DFS sites today are DraftKings and Fanduel, which control approximately 90% of the $3B DFS market.

There are three primary types of DFS games: Head-to-Heads (H2Hs), Double-Ups, and Guaranteed Prize Pools (GPPs). In H2H games, two contestants play for a single cash prize. In Double-Up games, a pool of contestants compete to place in the top 50% of lineups, which are awarded twice the entry fee. In GPPs, a pool of contestants compete for a fixed prize structure that tends to be very top heavy; some contests payout hundreds of thousands of dollars to the top finisher.

Over the last year, I have developed a winning system for daily fantasy football and baseball contests. Building this system from scratch was a fantastic compliment to the things I learned as a student, from machine learning and optimization to optimal learning and game theory. I hope others can join me in researching daily fantasy basketball and perhaps get involved with the burgeoning world of daily fantasy sports.

Content

This dataset contains 20 days of DraftKings NBA contest data scraped between 2017-11-27 and 2017-12-28. For DraftKings NBA daily fantasy basketball contest rules, see https://www.draftkings.com/help/rules/nba.

Format:

• One folder per day

• One folder per contest for a given day

• Salary file (“DKSalaries.csv”), payout structure file (“payout_structure.csv”), and contest results file (“contest-standings.csv”) for a given contest. Column headers in each files are pretty self-explanatory.

• Some additional files (e.g. “players.csv”, “covariancematunfiltered.csv”, “histfptsmat.csv”) for a given contest. These files were for my personal research, feel free to use or ignore.

• “projections” folder contains projections data for each player from rotogrinders and daily fantasy nerd, labeled by date.

• “contests.csv” contains information about each contest, e.g. entry fee, slate, and contest size.

Acknowledgements

Thank you to my friend from college, Michael Chiang, for contributions to this project.

Inspiration

A few ideas to get started:

• What kind of position "stacks" tend to maximize correlation within a lineup?

• How can you minimize correlation between lineups, such that you maximize your chances of winning a GPP?

• What are the tendencies of some of the top DFS pros?

• Can you improve rotogrinders and daily fantasy nerd player projections?

• Can you predict which players are undervalued (i.e. high fantasy points / salary ratio)?

• Can you predict the ownership percentage for each player in a given contest?