Anomaly Detection in Sports Training Data: An Improved Adaptive Algorithm

Identifying the anomalies in athletes' training data plays an important role in preventing injuries and optimizing training plans. In this paper, an improved adaptive algorithm is proposed to improve the accuracy and efficiency of anomaly detection in sports training data. Firstly, a depth self-encoder is constructed to learn the low-dimensional representation of sports training data, which captures the key features in the data. Then, the adaptive landmark filtering mechanism is introduced, and landmarks are selected and filtered and optimized based on the peak density to enhance the sensitivity of the algorithm to abnormal data. In the data preprocessing stage, the original data is cleaned and features are selected to reduce noise and extract the most relevant features. A composite loss function including reconstruction loss, affinity loss and sparsity loss is also proposed to further optimize the performance of the algorithm. Finally, in the training and optimization stage of the algorithm, the designed compound loss function is used to train the algorithm, and the parameters are adjusted and the algorithm is verified. The final results show that the improved adaptive algorithm shows the detection accuracy of more than 99% on multiple data, the missed detection rate remains between 0.257% and 0.447%, and the detection delay is lower than that of GAN and K-means algorithms, which shows the efficiency and accuracy of the algorithm in anomaly detection tasks. Therefore, the algorithm not only improves the accuracy of anomaly detection, but also enhances the applicability and effectiveness of the model in the actual sports training scene, showing a broad application prospect.