The aim of the thesis is analysis and evaluation of a basketball attack based on position of players on the court at a certain time. We develop an algorithm, that grades an attack with a value. We can split our thesis into two bigger parts. In the first part we develop an algorithm, that converts SportVU data into a reasonable timeline of a basketball game, and evaluate its results and problems. In the second part we first prepare the data for modeling and we describe attributes, models and different metrics for evaluating the models. For modeling we used a decision tree, random forest, logistic regression, support vector machine and artificial neural networks. We use these models to predict, whether the next action is going to be a shot, a pass or a turnover. Based on given results we chose the random forest as our main model. With the help of this model we compute several heuristics for evaluating a moment during an attack. For a better estimate we also build models for predicting direction of a pass and the probability of taking a successful shot. We evaluate these heuristics on an example attack and we compare the sum of all values given by the heuristics for every team and we compare them to the actual team standings.
|
