Improving Table Tennis Serve Tactics with Convolutional Neural Networks and Random Forests

This research study aims to analyze the table tennis serve shot tactics with the help of Convolutional Neural Networks (CNN) coupled with Random Forest (RF) models. The study aims thus to identify patterns and prescribe accurate serving strategies as seen through advanced algorithms. The current study employed a video database of table tennis professional players' match clips with high-quality videos. All the serves observed in the data set were described by information about the type of a serve (topspin, backspin, sidespin, no spin), speed, location, and outcome. Based on the existing literature, the CNN-RF model enhanced performance more significantly than conventional analytical methods. The model also exhibited better general performance results of about 89%, much higher than the 76% obtained through traditional techniques. The CNN-RF model had an accuracy of 0. * for both datasets, along with the precision, recall, and F1-score. 89 each, which highlights that consumers of the proposed approach also exhibit high accuracy in classifying or predicting all tiers of servers. The ROC curve plotted using binary logistic regression was 0. Substantial values of 92 imply that the created model has a sound discriminatory power, achieving the study's goal. The results of the training and validation measures were quite similar, showing the training accuracy of 94% and the validation accuracy fluctuating around 91% after the 20th epoch. The confusion matrix showed that the model did an excellent job correctly classifying service kinds, although mistakes were made.