An intelligent clustering framework for substitute recommendation and player selection

Player selection is an important aspect of team-based sports such as cricket. Various situations, like players getting injured, rested, or falling under disciplinary action, etc., are common in cricket, and in those circumstances, the proper substitution of players is very important. We present an innovative knowledge-based intelligent framework for substitute suggestions by employing various clustering techniques like DBSCAN, Spectral clustering. We compared it with the substitution made in the real-time team selection process and obtained a large similarity between that and the recommendations generated using Spectral clustering. We have also compared our proposed results with existing state-of-the-art works in our experimental setup, where they have used the K-means clustering technique. The results highlighted that Spectral clustering is the best choice for substitute recommendations among the mentioned clustering techniques. We also present an intelligent framework for team selection and apply various similarity measures like Euclidean distance, Cosine Similarity, Manhattan distance, and Pearson Correlation Coefficient to find the most accurate combination of players. The recommendations obtained from the Pearson Correlation Coefficient have a maximum accuracy of 77.50% with a high F-measure (0.77) and present different directions for the team line-up formation.

Automated summary

Player selection is crucial in team-based sports like cricket due to various situations that may require substitution. We propose a knowledge-based intelligent framework for substitute suggestions utilizing clustering techniques such as DBSCAN and Spectral clustering. Our results show a significant similarity between the suggestions generated using Spectral clustering and the actual substitutions made during real-time team selection. Additionally, our framework outperforms existing state-of-the-art works that use K-means clustering. Moreover, we present an intelligent framework for team selection using similarity measures like Euclidean distance, Cosine Similarity, Manhattan distance, and Pearson Correlation Coefficient, achieving a maximum accuracy of 77.50% and providing diverse directions for team line-up formation.