Cricket Team Selection Based on Complex Dynamics Using Machine Learning

Abstract

ABSTRACT Cricket is a sport which has a history, well established governing body and an economy built around. The selection of a winning team is a crucial measure, so the selection committees of teams take extensive efforts to formulate the winning team combination in terms of batsmen, bowlers and allrounders. Usually the selection process is meticulous and will be biased. Hence, this study mainly focused to propose a data driven approach where historic match data under complex dynamics being used and analyzed for an optimal team combination selection in cricket. The data for the study is secondary data scraped from cricket statistics repositories and data pertaining to the One Day International matches of the Sri Lankan cricket team since 2009. The data were feature engineered before conducting multi-output, multi-class classification tasks and Fuzzy Logic inferencing, and during feature engineering the pitch condition was derived using historic pitch report data. A neural network with 7 dense layers was designed which used 7 input features; stadium, stadium type, minimum and maximum temperature, average wind speed of the stadium, and the winner of each match and classify into three output the number of batsmen, bowlers and allrounders. The neural network yielded 76% of performance under 80:20 split on training and testing data for 100 epochs. Three Fuzzy Inference Systems were developed to rate and rank players based on historic performances. These inferencing systems have yielded 75%, 67% and 62% accuracies for Batting Performance FIS, Bowling Performance FIS and Allrounder Performance FIS respectively. This domain inherits data imbalance problem and there are unmeasurable attributes such as psychological, physiological and political aspects, deliverables of the study can be considered as a decision support models for cricket team selection. In future, more empirical methodologies need to be carried out to increase the performance of the neural network model and Fuzzy Logic Inference Systems in terms of a more adaptable model for squad selection in cricket.