This article focuses on the field of community health management for the elderly, and studies the demand for accurate prediction of the health status of the elderly. By collecting multi-source behavior data of 100 elderly people covering daily activities, sleep, social interaction, etc., an intelligent identification algorithm based on CNN (Convolutional Neural Network) and a HSPM (Health Status Prediction Model) based on LSTM (Long Short-Term Memory) are constructed. In the experimental stage, the model was trained and tested by dividing the training set (70%), verification set (15%) and test set (15%). The prediction accuracy of the model is about 60% at the beginning of training, and gradually increases with the progress of training, and finally stabilizes at about 85%. The loss function value decreased from about 1.2 at the initial stage of training to about 0.4. Compared with traditional SVM (Support Vector Machine) model and Naive Bayes model, this model has obvious advantages, with the highest accuracy of 70% for SVM model and 65% for Naive Bayes model. Repeated experiments show that the model is stable and the accuracy fluctuates between 83% and 89%. The research shows that the model and algorithm have excellent performance in predicting the health status of the elderly and can provide effective support for the health management of the elderly.
Chuman Luo (Sun,) studied this question.