Prediction of Future NBA Games’ Point Difference: A Statistical Modeling Approach

Abstract

Nowadays, the popularity of the NBA along with the penetration of gambling ideas is rising all around the world. More and more fans prefer to bet on sports gambling. People want to predict the final score difference for each individual game. In addition, there are many either superficial or underlying factors that will affect the game’s result. The central claim of this article is to establish a reliable model that could predict each game’s result by extracting and analyzing previous games’ outcomes. The major analysis is done by modeling and exploratory data analysis, and is composed of several graphical methods. For the modeling part, ideally the model could predict the games’ score differences by analyzing previous dozens of games’ data sets. The database consists of four seasons 2013-2017 on Goldsheet website to explore more analysis. Then step by step checking and demonstrating to prove the feasibility and precision of the model. The model will collect and analyze the data information, like teams, rebounds, assists, turnovers, three points, free throws, blocks, and injury data. Considering the transfer between teams and other changes, the model will predict the future game result in each season independently by using each team’s previous game result in that season. This article indicates the idea of linear regression model to find the best fit by comparing the correlation strength of each variable, which include the home-field advantage, teams’ technical statistics, and injury data, to the result of a game. The model final outcomes implies that the game’s result has the most correlation with the home-field advantage and player injury; however, team’s basic technical statistics, included the rebounds, blocks, turnovers, free throws, three points, and assists, have low correlation coefficient with the result to demonstrate that they are not significant to a game’s result.