Enhanced Sports Performance Analysis: CNNLSTM Based Detection of Tennis Serve Variations

This study looks at the applicability of a special network structure referred to as Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) in identifying minor changes in motion in the handling of the tennis racket during serves. The desired performance in rare movements, such as the incorporation of spatial feature extraction capabilities of CNNs and temporal sequence learning of LSTMs, attains serve. This model was trained and tested using a video-based dataset containing videos of professional/semi-professional and amateur tennis players. Next, a considerable number of small motion changes were extracted and labelled for many samples in the dataset. The observation made from the current research analysis shows that the CNN-LSTM model achieves an accuracy of 94%.5% on the detection of the disease with a precision of 92.3%, a recall of 93 - it is noticed that the above result is that which consolidates the fundamental constancy between the recall and the result it portrays. 8%, precision 93%, and F1-score 93%. But even in this regard, it stands at 0%, which is even higher than many state-of-the-art models. The analysis of the performance at each epoch shows gradual improvements; consequently, the training and validation sets attain high accuracy after 50 epochs. The optimization of the form of the batch size implies that an ideal batch size has been estimated as 32, with a corresponding classification accuracy estimate of 94 with a standard error of 0. Per cent: 5% and the decrease on 0. 25. The heatmap of the confusion matrix shows high classification rates for several of the motion variation classes with very little confusion, and the ROC curve analysis yields an average AUC of 0.93 for all classes.