Prediction model of physical activity level and diabetes mellitus based on artificial neural network

Objective: clinical practice shows that the control and treatment of chronic diseases, in addition to drugs, exercise health management is more significant, but also the cheapest, no side effects, thorough intervention. It has been proved many times that exercise can control and prevent chronic diseases. Through the related research of sports intervention, because the researcher selects the experimental object and the movement type difference, thus obtains the different research conclusion. Different types of people use different types of exercise, to achieve different exercise results, especially in the group of chronic patients. Therefore, to discuss the establishment of evaluation and selection system of exercise mode in chronic disease exercise management is the time requirement of health promotion for the aged patients. Methods 1242 patients with different degrees of diabetes in Henan Province were selected and their informed consent was obtained. Using the physical activity scale (IPAQ) and the long form questionnaire, the questionnaire was issued and collected from July 1,2020 to December 31,2020 by the uniformly trained physical examination physician, the data were analyzed by artificial neural network. Results: this study has found a certain correlation between diabetes and physical activity level whose data is incomplete, but has showed an overall positive correlation.


Introduction
Along with the development of the population ageing, the World Health Organization has developed the concept of healthy ageing, which emphasizes the individual health, physical health, mental health and good social adaptation of the elderly, the control and prevention of chronic diseases is one of the important connotations of healthy aging. In 2020, China's elderly population will rapidly increase to 260 million, until 2030,2050, China will enter the most severe stage of aging. At the same time, the number of people with chronic diseases is increasing with the aging of the population. China's chronic disease prevention and control plan (2012)(2013)(2014)(2015) shows that the number of people diagnosed with chronic diseases has reached 260 million, accounting for 85 percent of all deaths in the country. The prevalence of diabetes, for example, is still on the rise and is getting younger. Therefore, the World Health Organization stressed: sit less activity is the current frequency of chronic diseases of the first independent risk factors. At present, 88% of adults in our country do not exercise enough, "less exercise" The basic theory of LSTM network: The main structure of LSTM memory unit is gate and memory cell, gate also includes: input gate, output gate, forgetting gate. Gate structure can effectively retain and filter information, so memory cells can be maintained and updated.
Forgetting Gate takes the input information    The input gate uses the sigmoid activation function and the Tanh activation function to control the value of the new input information that can enter the cell state. Where the Tanh activation function generates new memory cells , the sigmoid activation function generates a weight between [0,1] that controls the amount of information entering the cell state. We can effectively combine an input gate with a forgetting gate to create a new cellular state The output gate uses the Tanh activation function to process the current cell state

Logical analysis
Comprehensive Questionnaire Survey and research results, combined with Literature Review, logical analysis of the problems and phenomena, concluded the research conclusions.  Table 1 clearly shows that there is a significant correlation between the level of physical activity and diabetes. The smaller the Loss value, the closer to zero, the better the model's predictive ability.

ROC of the predictive model for predicting the risk of diabetes by the level of physical activity
When the artificial neural network is calculated, it is divided into diabetic and non-diabetic patients, so the roc curve is used to describe the prediction result. For the true positive rate on the ordinate and the true negative rate on the abscissa, the closer the enclosed area is to 1, the better the prediction result. The ROC value of this model is 0.92, which shows that the prediction effect is better.

Discussions
The regression analysis of the different physical activity levels of diabetic and non-diabetic people shows that the level of physical activity is highly correlated with the prevalence of diabetes. At the same time, through the deep learning and training of the artificial neural network, it is known that the level of physical activity and the diabetes risk prediction model are highly reliable. This conclusion provides a certain reference for the clinical prediction and management of diabetic patients, and reminds people to pay attention to the problem of disease prediction. At the same time, this study still has certain shortcomings, such as the age difference of patients and the body composition has not been analysed, and further research is needed. The authors would like to thank for these financial supports.