Anomaly Detection and Early Warning Model for Physical Fitness Monitoring Data based on Deep Learning
Xinyu Chen, Song Li · 2025
If the outliers in the physical fitness monitoring data are not handled in time, they may mislead health assessment and training plan formulation. This paper collects physical fitness monitoring data from various wearable devices and professional physical fitness testing instruments and preprocesses the collected physical fitness monitoring data. Subsequently, CNN is used to extract features from the preprocessed data and extract local features in the data. On this basis, the features extracted by CNN are used as input, and LSTM is used for time series modeling. Next, an anomaly detection module is designed on the output of LSTM to determine whether the data is abnormal. Once abnormal data is detected, the early warning mechanism is triggered immediately. The model proposed in this paper performs well in the accuracy of anomaly detection of physical fitness monitoring data. The accuracy of most data points exceeds 95%, and even individual data points reach 100%. In addition, the warning delay of the model is significantly lower than that of naive Bayes, with an average warning delay of 303ms. This paper provides reliable data support for health assessment and training plan formulation.