Neural Network Forecasting of ”Quality of the Day” for the Category of People Sensitive to Weather with Cardiovascular Diseases

With the development of neural networks and machine learning, intelligent forecasting is increasingly used in the problems of predicting the risk of occurrence and development of socially significant diseases, the first place among which is occupied by cardiovascular diseases (CVD). A large group is the category of weather-dependent people, for whom the risk of occurrence and development of CVD is correlated with meteorological factors and solar activity. The construction of effective predictive models of the ”quality of the day” for weatherdependent people allows to increase the efficiency of taking preventive measures and the quality of CVD treatment. In this research paper, the issues of using recurrent neural networks (RNN) of various types for forecasting multivariate time series in relation to the problem of predicting the ”quality of the day” for weather-dependent people for CVD are studied. The frequency of admission of patients with CVD to medical institutions served as an evaluation parameter for determining the ”quality of the day”. The input parameters for forecasting were ambient temperature, air humidity, atmospheric pressure, geomagnetic activity index and the rate of their daily change. Forecasting was performed using the sliding time window method using MLP, RBF and LSTM neural networks. The ”quality of the day” was assessed based on the Mamdani fuzzy model. Analysis of the obtained results shows that the accuracy of RNN forecasting is not stable, both in time and in many different types of neural networks. This leads to the conclusion that it is advisable to use ensemble modeling in problems of forecasting the risk of occurrence and development of CVD.