Innovative Innings: Exploring Cricket Shot Categories with CNN-SVM Methodologies

Automatic identification of cricket strokes is essential for preparation and training in sports analytics. A research paper presents an innovative approach to cricket shot classification. It combines convolutional neural networks (CNN) and support vector machines (SVM) to classify five specific types of cricket shots: cover drive, cut shot, pull shot, leg glaze, and straight drive. The model's performance is evaluated holistically using numerous criteria such as precision, recall, F1-score, support, support percentage, and overall accuracy. The model showed perfect accuracy, ranging from 86.91% for the cover drive to 94.29% for the straight drive, reflecting the model's ability to detect positive occurrences of each shot type accurately. The recall rates were outstanding, and the maximum rate was 97.16% for Pull Shot, which reflects the model's sensitivity in capturing the critical shots. All classes produced consistent F1 scores, which measure the balance of precision and recall. The highest of 94.38% was achieved for Pull Shot, confirming the excellent evaluation of the model's accuracy. The support proportions were computed using the support values, which provide the absolute occurrences in each class, showing the class distribution across the dataset. Each class had a support ratio ranging from about 0.19 to 0.22, with the Straight Drive class having the highest percentage of 0.22. The model obtained an excellent accuracy of 91.08%, proving its utility in all the classes. In addition to specific indicators, we summarized the general performance of the model using a few averages. The macro average that puts equal weight on all classes yielded an accuracy of 91.07%, recall of 91.35%, and F1-score of 91.11 %. Considering the class imbalance, the weighted average gave an accuracy of 91.21 %, recall of 91.08%, and F1-score of 91.04%.