Using Convolutional Neural Network and Candlestick Representation to Predict Sports Match Outcomes

The interdisciplinary nature of sports and the presence various systemic non-systemic factors introduce challenges in predicting match outcomes using a single disciplinary approach. In contrast to previous studies that use performance metrics statistical models, this study is first apply deep learning approach financial time series modeling predict outcomes. proposed has two main components: convolutional neural network (CNN) classifier for implicit pattern recognition logistic regression model outcome judgment. First, raw data used prediction are derived from betting market odds actual scores each game, which transformed into candlesticks. Second, CNN classify candlesticks on graphical basis. To end, original 1D encoded 2D matrix images Gramian angular field then fed classifier. way, winning probability matchup team can be based historically implied behavioral patterns. Third, further consider differences between strong weak teams, adjusts by makes final judgment regarding outcome. We empirically test 18,944 National Football League game spanning 32 years find individual historical better than all teams. conjunction with outperforms SVM, Naïve Bayes, Adaboost, J48, random forest, its accuracy surpasses prediction.