Predicting of sports events results

Abstract

Today, the field of betting and bookmaking is popular with a wide range of sports fans. Issues of predicting the outcome of future events are and will be relevant for everyday life, sports, politics, etc. With the increasing number and quality of methods of intellectual analysis, the idea of predicting the results of sporting events became feasible. Applying different mathematical methods helps to obtain more accurate predictions of results than subjective expert estimates. The paper introduces the concept of betting and describes in general terms the task of bookmaking. The purpose of the study and the tasks that must be accomplished to achieve the goal are identified. Existing research results of different scientists who have researched this problem are analyzed. There are four basic principles for predicting the outcome of sports events. Different approaches to the task have been considered and our own way of solving it has been proposed. Methods such as Poisson distribution, simulation modeling of the Markov Monte Carlo chain, and many other research methods have been considered. The formulation of the problem is formulated and the properties of the problem are investigated. A backtesting algorithm was developed and described as a mechanism for presenting team statistics at any point in time for a particular season to collect sports event data. Correlation analysis for the selected parameters was shown to show a moderate correlation of data and the use of Google AutoML to identify patterns between the data was described. The importance of using machine learning to solve this problem is outlined. A system has been developed that collects event data and calculates statistics for each team at each point of time using the backtesting algorithm. A service has been developed to create and test the quality of the strategy. The results of experimental studies of task efficiency are presented, where we conducted experimental sets of strategies with and without adding the result of the AutoML service and for each strategy the Pearson correlation coefficient was calculated based on the results of two past seasons. The results obtained are analyzed.