Can you predict an NBA player's performance during a game? Technion did

Researchers at the Technion developed a new method to predict NBA player performance based off of past performance metrics and pregame interviews.

Portland Trail Blazers shooting guard Sasha Pavlovic (3) drives against Golden State Warriors point guard Jarrett Jack (2) during the first half of their NBA basketball game in Portland, Oregon, April 17, 2013 (photo credit: STEVE DIPAOLA/REUTERS)
Portland Trail Blazers shooting guard Sasha Pavlovic (3) drives against Golden State Warriors point guard Jarrett Jack (2) during the first half of their NBA basketball game in Portland, Oregon, April 17, 2013
(photo credit: STEVE DIPAOLA/REUTERS)
Is it possible to predict the performance an athlete will have over the course of a game? Technion researchers seem to have found out how.
To do so, the researchers developed a new computational method for predicting the performance of professional basketball players by anyalyzing the player's profile, which includes past performances as well as pregame interviews.
The method itself was developed by doctoral students Amir Feder and Nadav Oved, under the supervision of Professor Roi Reichart of the William Davidson Faculty of Industrial Engineering & Management. Their findings were published in the journal Computational Linguistics.
Today, prediction mainly relies on a small ranges of data, which is mostly based off of a player's past performances.
The researchers, however, discovered "out-of-game" factors that are important in determining a player's upcoming performance, such as the transcript of pregame interviews, which the researchers purport contain small cues that can improve predictions about a player’s behavior and their possible performance in an upcoming game.
Many external parameters also needs to be taken into consideration, such as the environment of the game, decision-making, rational as well as the disposition of the player - given that the activity takes place in a complex and dynamic space.
According to the study, the researchers developed several models, utilizing neural indicators to predict the actions of the players based on what they said during their pregame open-ended interviews. The models are capable of making predictions based off of interview text alone, or a combination of interview text and past-performance metrics.
The published study utilized 5,226 pairings of interviews and past performances for 36 outstanding NBA players to compile their dataset. The researchers matched this up against in-game performance metrics following the interviews, correlating their performance and demeanor during their pregame interview to determine if it had a linking effect on their play.
Each pair was evaluated by correlating the transcript of the interview with performance indicators during the game, such as risk-taking characteristics, a player's behavior and decision-making abilities. For example, a risk can be defined as an attempt to make a three-pointer in poor positioning - choosing to take a defense approach to the game is an example of behavior or strategy.
NBA player performance prediction accuracy. Columns from left to right: Dataset majority baseline - naive prediction method; Metric-only baseline - prediction based on past performance only; prediction based on interviews (method developed by Technion researchers); prediction on interviews and past performance.NBA player performance prediction accuracy. Columns from left to right: Dataset majority baseline - naive prediction method; Metric-only baseline - prediction based on past performance only; prediction based on interviews (method developed by Technion researchers); prediction on interviews and past performance.

The results of the study showed that the text-based models outperformed strong baselines normally working off of performance metrics alone - demonstrating the importance the pregame interviews for actionable predictions. The researchers determined that models using both interview and past performance metrics improved on some of the most challenging aspects of predicting player performance, and in turn produced the best results.
 Prediction accuracy of the model per player, relative to its accuracy for all players (black line), for each prediction task. Points to the right indicate better than average prediction. (photo credit: TECHNION) Prediction accuracy of the model per player, relative to its accuracy for all players (black line), for each prediction task. Points to the right indicate better than average prediction. (photo credit: TECHNION)
For example, according to the report, during a pregame interview  the 2016 NBA Finals, LeBron James, then with the Cleveland Cavaliers, was asked by reporters about his mental state and how he was feeling based on his personal history and previous attempts to bring home the title for Cleveland (James was born in Cleveland, and returned to the team to bring its first championship). Within the interview, James described his positive mental state, concentration and feelings of ease going into the games.
The Cavaliers successfully brought home the Larry O'Brien Championship Trophy that year - the first for the team. James was named MVP of the series.
Accordingly, Prof. Reichart explained, "Our models processed the text and guessed that James' offensive performance would be better than his past averages. In practice, the 2016 Finals series ended with Cleveland's first - and only - winning championship. In these games, James surpassed himself and starred throughout the series, as our models predicted."