Classification Algorithm And Design Of Application System Based On Personal Activity Date

In recent years,with the development of the Internet of Things technology and the popularization of smart phones,from the perspective of the development of the smart healthcare industry,wearable devices have huge potential.With the advancement of sensor technology,users can use wearable devices to accurately monitor their sleep status and daily exercise in real time,which can be used for various health care and preventive medical treatments.The social health care system is closely related to people's lives,and current medical diagnosis mainly relies on physical hospitals,and there is a greater demand for preventive health care,which makes wearable devices have broad market prospects.As the basis of personal health testing in the future,wearable devices will occupy an important position in personal health management,chronic disease management,personal disease prediction and medical diagnosis.This article mainly studies abnormal detection and health prediction based on personal exercise data.Algorithm,including the following:First of all,the traditional method of density estimation will cause a phenomenon of "dimension disaster",which leads to the problem of poor detection effect.The Gaussian Mixture Generative Model(GMGM)is designed.GMGM uses a variational autoencoder to train the original personal motion data,and extracts the potential features of the data by reducing the reconstruction error.Because the traditional two-step detection technology will lose key information and reduce the detection accuracy when performing anomaly detection,GMGM jointly optimizes the variational autoencoder,deep belief network and Gaussian mixture model in an end-to-end manner,retaining the original data feature.Experimental results show that the proposed method can effectively improve the anomaly detection effect of personal motion data.Secondly,in order to solve the problem of a lot of available human activity data but lack of disease tags,an unsupervised representation learning model tim2 vec is proposed.This method learns the feature representation of time series data from the original activity data,and can be used for health by mining through distributed representation.The activity mode of the situation prediction task.By considering the periodicity of the activity level,embedding the ordinal relationship representing the activity level,using noise contrast estimation to construct a learning loss function.Finally,the learned features are introduced to construct a stacked two-way GRU model for health prediction.Experiments on 2 data sets show the effectiveness of the proposed method.Finally,based on the B/S architecture,starting from the first two theoretical research points,construct a health prediction and personalized health recommendation system.Based on personal basic information,combined with the user's feedback on personalized health advice to construct a user portrait,a collaborative filtering algorithm based on correlation coefficients is used to calculate the S-TF weight between items,and the result with the highest score is selected for output,providing users with Personalized health advice.Next,the graphical interface of each functional module is shown,and the advantages of this system are demonstrated through functional comparison with common sports and health APPs on the market.