Diabetes Prevalence Forecasting Using GARCH and Deep Learning Models

Abstract

The aim of this study is to use cutting-edge statistical and artificial intelligence techniques to assess and forecast the prevalence of type 1 diabetes in Iraqi children. In order to determine the prospective ultimate caseload and the fictitious inflection point, the research used the generalised autoregressive conditional variance (GARCH(p,q)) model to estimate the temporal behaviour of the data. The models under test were subjected to statistical stability conditions, and the AIC and BIC information criteria were used to assess their performance. The findings indicated that, out of all the nonparametric models examined, the GARCH(1,0) model performed the best. However, the paper showed that there were no significant variance fluctuations in the incidence data, which restricts the use of strictly nonparametric models. Deep LSTMNs neural networks were used for time series analysis in order to get around this. Using the ideal parameters for the number of layers and iterations, the data was divided into an 80% training set and a 20% test set. In comparison to other models, the LSTMNs findings showed the lowest mean squared errors, indicating a good capacity to predict future values. This implies that an efficient method for tracking upcoming changes in chronic illnesses is to combine deep learning algorithms with traditional statistical models. To enhance planning and response to the epidemic load of chronic illnesses, the paper suggests using this strategy in medical modelling and health forecasting.