The NBA MVP race is usally an exciting race to watch. Being awarded as the NBA MVP for a season is the highest inidividual honor a player can receive. As Basketball Joe points out, every NBA MVP has been inducted into the Hall of Fame.
The objective of this project is to predict who will be awarded the MVP of the 2019-2020 season using past historical data regarding MVP. This will be determined by looking at the player who has the most amount of votes for the season in reference.
I would frame this as regression problem as I am looking to predict the shares of votes each player that played in the season will receive. Can also think of this as a an offline, model-based, supervised (regression) learning system.
To that end, my main performance measures will be performance measures commonly used for regression problems: RMSE and MAE
My guy Basketball Joe has a great article about historical MVP voting criteria that I will be using to help guide this project. In his article, he states that there are three branches of MVP criteria: team success, individual statistics, and media narratives. His article provides some key pointers as to what may be the most important features in NBA Voting.
Right before the 1980-81 season the NBA changed its voting method and allowed sportwriters and broadcasters to vote..
66% of the league's previous MVPs where awarded from the teams with the best record and 84% of the previous MVPs were awarded from the teams with the two best records. What happens when a team has multiple all stars? Does that hurt these players chances of winning?
When a team achieves the best record in the league with just one All-Star, the choice for MVP is clear. When a team achieves the best record in the league with 3 or 4 All-Stars, the choice for MVP is not so clear.
For the most robust model, I think I need all the players that played in NBA games from the 1980-81 season to the 2018-19 season. The NBA API allows me to get all career stats for all players. I can the augment this with data from BasketballReference.com.
I will use the players from the 2019-20 season as my test data set.