Rallying Innovations: CNN-SVM Frameworks for Badminton Shot Recognition
Satvik Vats, Shiva Mehta · 2024
This study presents a technique based on a combination of a Convolutional Neural Network (CNN) with a Support Vector Machine (SVM) model for well-honed recognition of player shots in badminton, designed to help collect statistical information. Our developed model underwent a long training process with the dataset consisting of 1990 photos; this gave the model a general accuracy of $92.686 \%$. The specialized study for each class exhibited a wide range of accuracy rates, starting with $89.21 \%$ for Class 1 and ending with 95.34 % for Class 5. Recall rates have also shown much variety, such as 86.16 % for Class 4 and 97.7 % for Class 3. The F1 scores fluctuate between 69.06 % and $85.78 \%$ for Class 3, which shows the model can identify the shots correctly, accurately, and consistently. Micro, modified, and weighted average accuracy prominently above $90 \%$, a value measure, was used to assess a model’s performance. This thoroughly holistic approach will allow the model’s effectiveness to be evaluated fully. The scores to indicate the faithfulness and evenhandedness of the model’s predictions in multiple categories are only a few examples. The overall macro averages, representing the class balanced goodness in the model’s performance, for accuracy was 92.66 %, $92.89 \%$ for recall, and $92.7 \%$ for the $F 1$-score. In the second approach, the weighted averages showing the course distribution were next to each other in the range, allowing us to explore the model’s efficacy. The micro-precision found that the model achieved an accuracy of $92.69 \%$ in every scenario. Moreover, CNN for feature extraction and SVM for classification created an outstanding prospect that later revolutionized the field of sports analytics, specifically the strategic area of badminton shot categorization.