ANALYSIS OF MACHINE LEARNING MODELS PREDICTING BASKETBALL SHOT SUCCESS
by Max Murakami-Moses
Category: STEM
Abstract – The most critical aspect of winning a basketball game is shot selection. However, due to the multitude of factors that come into play when deciding if a shot was a good shot selection (a shot that has a high chance of going in is a good shot selection), it is difficult for a human to make a reasonable assumption. Because of this, we used and analyzed a variety of machine learning techniques to predict shot success.
Here, we perform an analysis of the best machine learning models for predicting shot success as well as comparing the performance of these models using different features. Our models (neural network, logistic regression, and gradient boosting) were able to predict shot success between 64.9% - 65.1% accuracy.
Our models performed with about equal accuracy; however, when we altered the features, the accuracy significantly decreased. This highlighted the importance of good quality data as well as the importance of certain features, such as the type of shot, in making a shot.
Here, we perform an analysis of the best machine learning models for predicting shot success as well as comparing the performance of these models using different features. Our models (neural network, logistic regression, and gradient boosting) were able to predict shot success between 64.9% - 65.1% accuracy.
Our models performed with about equal accuracy; however, when we altered the features, the accuracy significantly decreased. This highlighted the importance of good quality data as well as the importance of certain features, such as the type of shot, in making a shot.