Efficacy of CNN-LSTM Structures in the Identification of Subtle Motion Variations in Tennis Serve

This research study will dive deep into a hybrid Convolutional Neural Network (CNN) model along with Long Short-Term Memory (LSTM) that is used to categorize the serves of a tennis player. Identifying the motion of the serve with precision and accuracy is one of the valuable factors in this game, and it has the potential to give great insight into training and performance. The proposed model based on CNN's prominent geographical feature reveals the power and ability of the LSTM to carry out a deep analysis of the temporal correlations, which is the most effective means to identify different types of serves. Implementation of the model was empirically evaluated on five classes of serves, which showed a slight improvement in both micro, macro, and weighted average of the accuracy, recall, and Fl-score. All those metrics, over 90%, went beyond. The statistical findings provided evidence that the model is so strong. The remarkable values of accuracy, recall, and Fl-score macro averages with 92.07, 92.42, and 92.08, respectively, can also be used as evidence. The values for averaging pondered by precision, recall, and Fl-score are 92.57, 92.15, and 92.2, respectively. Another factor is the so-called micro average of 92.15 for the three metrics. The model was constantly giving high accuracy, evidenced by staying below 0.96 for all classes. In the future, researchers will focus on strengthening the model by increasing the data size, including amateurs.