This is a review of the basketball research conducted by Min-hwan Oh, Suraj Keshri, and Garud Iyengar applying probabilistic graphical models for basketball match simulation.
With any sporting event, it is natural for analysts, bettors, and fans to make predictions regarding the outcome. This is certainly true within the National Basketball Association.
A simulation infrastructure is developed to bring together player identity and team level network. A basketball game is modelled using a probabilistic graphical model, which illustrates every touch and event during a game as a sequence of transitions between discrete states. The progression of a game is treated as a graph in which each node is a network structure of players, the actions, and events while the edges illustrate possible moves in the flow of the game. The conditional probability of the edges is learned while ball movements between players, how likely a player is to take shot, and how defense and teammates affect the dynamics of the game are all simulated.
The start of a possession is modelled as a multinomial distribution between players on the court. The probability of a field goal attempt is modelled as a Bernoulli distribution based on the idea that the likelihood of a player taking a shot depends on his tendency to shoot, his defender, and also the tendencies of his teammates. Shot efficiency is modelled as a function of the offensive player’s skill, the defender at the time of the shot, and the location of the shot. Passes between players are modelled as a network. Shooting fouls are modelled as a function of the shooter’s skill at drawing a foul, the defender’s foul tendencies, and the location of the shot. A free throw percentage for each player is used to sample a free throw success event. Rebounds are modelled as a competition between the players on the court. Possession start begins with an inbound pass, a defensive rebound, or a steal. Two types of turnovers are considered, stolen balls and all other turnovers that result in an inbound pass. The average probability of turnover per touch for each player is calculated from historical data.
The model is used to simulate the 2013-14 season record for each NBA team resulting in good estimates of the teams’ actual win percentages. For the simulation, the lineup is the input parameter. Matches are simulated multiple times to estimate the expected statistics for both players and teams. Results indicate that changes in a team’s lineup or the opponent’s lineup has a significant impact on the dynamics of how the game progresses.
The model can be used to evaluate the effect different lineups have on game results. Coaches can evaluate hypothetical lineups against an opponent in order to determine which lineup will be the most effective. Coaches can also model hypothetical opponent lineups in order to work out the best defensive strategy. Specific player performance can also be evaluated. This will help when looking at making trades and determining which players would bring the best improvement to the current lineup.
Analytics Used: Probabilistic Graphical Model, Multinomial Distribution, Bernoulli Distribution
Find out how Sports Analytics Expert Victor Holman can give your team the competitive advantage.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Discover the Groundbreaking Sports Analytics Application and Framework coaches and sports analysts are talking about!
Learn all about sports analytics in Victor Holman’s Sports Analytics Blog.