Predicting and explaining match results through statistical models and artificial intelligence techniques : a scientific approach to winning more

Abstract

The use of artificially intelligent techniques and statistical models are becoming increasingly popular in today’s world - especially in the entertainment industry. Gambling houses and online-based betting portals strive to provide the best possible odds in order to attract more clientele. As the gambling industry exponentially grows, companies are looking into modern approaches on how to predict results before a match has even started. A popular sport that is often attempted to be predicted is football, or as it is widely known in North America - soccer. Apart from having employees specialising in mathematics and statistics, such companies are now also investing in personnel that are solely focusing on machine learning and artificial intelligence techniques. Punters are often only presented with a static value dictating the possibility of a particular event happening. Furthermore, there are no platforms that provide databacked research, statistics and predictions for fixtures that are yet to be played. In this research, a Poisson distribution to serve as an accuracy baseline was developed. Additionally, a RandomForest and Deep Neural Network were implemented to outperform the aforementioned accuracy baseline. To achieve this, an extensive data collection and engineering process was required to craft the required datasets. After generating the required datasets, the predictions were trialled. The Poisson distribution did not generate a positive return on investment. However, the features generated from this distribution proved to be very useful when training machine learning models. In contrast, the ML-based algorithms outperformed the Poisson distribution from both an accuracy and financial return perspective. Accuracy rates did not outperform previously explored publications, however, this implementation managed to surpass return on investment by roughly 4% which was satisfactory. The implemented system guides users to place wagers that are data-backed, instead of using their personal instinct.