Deep Learning using CNNs for Ball-by-Ball Outcome Classification in Sports

In this paper, we compare the performance of three different Convolutional Neural Network architectures, inspired by literature on activity recognition in videos, on the novel task of classifying ball-by-ball outcomes for sports videos, specifically for cricket. In the sport of cricket, each ball can have one of four different outcomes, and we define our classification task as the prediction of this outcome, given a sequence of frames representing the video for that ball. We report the performance of each of these architectures, and explore their advantages and disadvantages for this domain. Our best model is able to achieve an accuracy of almost 80% on our validation set.