Using Convolutional Neural Network to Predict Sports

Convolutional Neural Networks, powerful Machine Learning tools for image classifications, can also be employed to perform sports outcome predictions. To use a Convolutional Neural Network for sports predictions, we arranged the statistics of the two opposing teams or players into a grid representation. We have exploited this two-dimensional input arrangement to expand the training set for our Convolutional Neural Network. The expansion consisted of shifting each grid that represented a game by one and two columns to the right. This shifting idea made it possible to employ Convolutional Neural Networks in predicting sports events without relying on extensive historical data. We used the Men Euro 2020 and Women US Open 2021 as test cases to illustrate this approach. The most performant models from our exploration registered a 70.2% accuracy in predicting the Women US Open 2021 and a 69.8% accuracy in predicting the Men Euro 2020. These accuracies are considered as improvements. The ensemble techniques we previously used on these datasets had an accuracy of 64.9% on the Women US Open 2021 and 67% on the Men Euro 2020.