Performance Forecast and Analytic for Sports Dance Athletes Using a Hybrid Method

This paper proposed a forecast method consisting of Long Short-Term Memory (LSTM) network with Markov transition matrix aiming for the forecast of training performance for dance athletes. Firstly, using the Event-Group theory to design five training indicators affecting the training performance of dance athletes. The role of the five training indicators is the construction of a training dataset. Thereafter, we put Markov transition matrix into LSTM network; meanwhile, we established the mapping of Markov transition matrix to LSTM network through writing the m-step Markov status matrix into the forget gate weight of LSTM network. Finally, using the experiments to verify the proposed method, and results show that the proposed method obtains 0.972 accuracy in forecasting the training performance of dance athletes and significantly outperformed the comparative methods in forecast capabilities. Results also show that the running efficiency of the proposed method defeated most comparative methods. Moreover, we find that the five training metrics can be used separately in the training of dance athletes, thus significantly improving their training performance, due to they exist weak dependency relationship. We also find that basic posture training has more positive effects than speed training in the improvement of the training performance for dance athletes.