Reliable Predictive Analytics for Sri Lankan Cricket in ODIs

Abstract

Cricket is one of the fascinating team sports that attracts millions of spectators world- wide. In the modern day cricketing arena, teams have to maximize both individual and team performances to gain a competitive advantage over other teams. Selecting most suitable players from a pool for an upcoming tournament is an essential task to con- sistently win matches. Team of selectors face di culties when doing this task as the task of selection is subjective and uncertain. Moreover, identifying all-rounder players is bene cial for any cricket team since they are indeed a better resource to the team as they are considered as one player with the skills of two players. Unfortunately, there is no formal de nition for identifying all-rounders in global or local context. Considering aforementioned issues, and after examining approaches proposed in the exis- tent literature, it was found that predictive analytic techniques can be conveniently used for this purpose. Especially, a fuzzy logic based methodology can eliminate the complex- ities and the uncertainties that occur between distinguishing player performance levels and make the selections more convenient and reliable. The batting and bowling per- formance fuzzy score results can be extended to reveal signi cant statistical facts about performance spread of all players. Additionally, proposed methodology for all-rounder classi cation utilizes the results (fuzzy scores) of player performances obtained in the player performance evaluation. Initially, a reliable data set was constructed by scraping and preprocessing data from ESPNCricinfo web site. Thereafter, based on the extracted features such as overall performance, current performance and rst class performance of individual players, fuzzy inference systems were developed to generate fuzzy scores for each player under three categories; batting, bowling and wicket keeping performances. Each player was then ranked based on these three categories. To select the best 15 squad for a given series, highest ranking players from each performance list was selected according to a given ratio. Moreover, by plotting the generated batting and bowling performance fuzzy scores, a statistical method was employed to categorize each player into one of the four categories; genuine all-rounders, under-performers, specialized batsmen and specialized bowlers. The best 15 player selection module was evaluated by comparing the squads generated by FIS with the actual squads selected by SLC Board for few past series. Experimented results were promising and indicated the e ectiveness of our approach. Moreover, all- rounder identi cation module provided an acceptable classi cation.