Predictive analysis of successful basketball shots?: The Euroleague case

Official URL: http://risc01.sabanciuniv.edu/record=b1857583 (Table of Contents)

Basketball industry creates vast amounts of data from which the organizations benefit to improve their business processes like revenue management, roster selection, fan engagement, and on-field decision making. Sophisticated data collection systems are being developed in order to get the maximum benefit from analysis of the movements and actions of all elements in the field. Since elite teams do not have huge differences when compared to each other in terms of advanced fundamental, physical capacity, and motivation, this valuable information is helping them to develop data driven decision making applications which give them a significant advantage on winning. In this thesis, we analyze ten seasons of the Euroleague professional basketball data consisting of spatiotemporal, player based, and situational variables such as score difference, shot type, and home or away team. Using these variables, we build predictive models for the accurate prediction of successful shots. We develop binary classification methods such as logistic regression, random forest, naive bayes, support vector machines, and artificial neural networks. We compare these models to evaluate the best approach for classification problems of successful basketball shots. Among all models we applied, random forest is the most accurate and logistic regression is computationally the most efficient model.