Generic model for automated player selection for cricket teams using recurrent neural networks

Abstract

In this paper, an optimized model has been proposed exclusively for the game of cricket whereby a team of fifteen members can be chosen in an unbiased strategy. The proposed method involves a hybrid technique that uses the concept of genetic algorithm (GA) and recurrent neural networks (RNN) for selecting efficient players. Suitable preprocessing is applied to the individual players historical statistics and an initial feature matrix is generated for a player. This feature matrix is fed to the proposed mathematical function used in GA. The GA utilizes a novel fitness function for the minimization of loss-factor. This results in a refined feature matrix. This refined feature matrix is further subjected to RNN to assess a final score for the individual player. Finally, the proposed model comes up with a concurrent rank table that can be referred by the team selectors for an easy and efficient player selection for the upcoming match. It may be essentially considered that the proposed model provides results for the upcoming match or tournament only. Three different authentic datasets have been referred for the purpose of experimental evaluation of the proposed model. The results are compared with the recent match performances of each players. Surprisingly, almost equivalent results are obtained which supports the robustness of the scheme. In cases, the proposed scheme outperforms the manual team selection in terms of performance, that indicates that a slight better team could have been created in certain cases. The overall rate of accuracy in terms of predicted list of players for each match vis-a-vis the manually selected players comes out to be satisfactorily 98.5%.