Detection of Match-Fixing in Football Matches Using a Conformal Anomaly Detector

Анотація

A complex problem that threatens the integrity and authority of football in many countries of the world, including Ukraine, is fixed matches, as they are also called – matches with a fixed result. The results of fixed matches, related to the winning of bets, can be considered atypical, or abnormal, which allows formalization of a search for such matches. To check the current match for a fixed result, mathematical methods of football analytics, such as prediction of the match result, and analysis of bets or actions of the match participants throughout the game, are used. Their advantage is the speed of decision-making, and the disadvantage is the need to use a huge amount of data, that is not publicly available. An approach when the decision about the fixedness of the match is made after the end of the season, based on the results of the games played by all teams, can be considered as an alternative. This approach allows to formalize the search of matches, suspicious for a fixed result, as the detection of contextual anomalies. Statistical non-parametric histogram methods are the most adequate for the considered task of identifying suspicious for a fixed result matches, according to the results of the whole season. However, for effective use, these methods require a significant volume of the sample, which is not performed for the considered task. A new method of finding anomalies in data is a conformal anomaly detector. It does not require knowledge of the distribution laws of the input data and also allows entering estimates of guaranteed accuracy for the obtained solutions. A method of detecting suspects for a fixed result of football matches based on the results of the entire season, using a conformal anomaly detector, has been developed. To evaluate the effectiveness, main classification metrics were used: precision, recall, and F 1 metrics. The peculiarities of using the method, based on the conformal anomaly detector, according to the data of individual classes of the model season are considered. A comparative analysis of the developed and histogram methods was carried out based on the data of the model season. Proposed detection method based on conformal anomaly detector provides a gain in detecting potentially suspicious fixed-score matches compared to the known histogram method by 13%-17% in the precision metric, 13%-21% in by the recall metric and 0.15- 0.23 by the F1 metric.

Класифікація

Ідентифікатори

Схожі роботи