Application of CNN-RF Models for Enhanced Serve Shot Classification in Table Tennis

In this research, a novel CNN-RF model is proposed to enhance serve shot strategies in table tennis. Combination of sift, detect and segment capability of CNNs with the maximum accurate classification capacity of RFs help to assess serving characters including the trajectory of the ball, direction of spin and movement patterns of players. This proposed model achieved an average accuracy of 92.5% which is quite higher than classic CNN-only (87.3%) and RF-only (82.1%).Using k=5 of Cross-validation, it was eliciting a standard deviation of ±1.2% this shows how stable the model is after partitioning the data. The model demonstrated good predictive capabilities: The accuracy of the model was at 91.8% and the recall was at 93.1%, and the F1 score at 92.4%. A feature relevance analysis done revealed that ball spin direction, serve speed and player movement explained serve success at 35%, 27 % and 20% respectively. The model was additionally tested with a confusion matrix, it classified topspin serves at 95% accuracy, backspin serves at 90%, and no-spin serves at 86%. These results support the ability of the model to correctly categorize between different serve kinds with fairly high accuracy. The hybrid CNN-RF model has large practical implications that sample actionable information for improving training programs and real-time service ways. Thus, the finding of this study is that the CNN-RF model can be regarded as a more accurate and consistent means of serve shot analysis in comparison with traditional methods; moreover, the method has the most prospects for the development of other sports' metrics assessment.