Statistical Prediction of Diabetes Prevalence in Iraq using a Hybrid ARIMA–LSTM Model

Abstract

This study concerns the emerging challenge of diabetes in Iraq and aims to forecast its prevalence with advanced statistical methodologies . Diabetes is a rising health problem in Iraq attributed to variables from obesity to hypertension and an outdated healthcare system. Conventional linear models fail to well illustrate the complicated and non-linear nature of the history of the disease, thus we use a two-stage method in this work. The linear trend is extracted from the data with ARIMA and then it is used to predict for non-linear analysis using LSTM. Data were gathered, refined and analyzed from 7 Iraqi governorates. The diabetes prevalence decline identified by using the ARIMA-LSTM hybrid model in this data suggests that the integrated model showed better prediction of diabetes prevalence than individual BPMs. The hybrid model was found to be the best way to reduce the prediction error, noise and incomplete data processing. Given the robustness of the model in various locations and sample sizes, it could be considered as a decision support instrument for health policy planning, early detection strategies, prevention and resource allocation. The paper suggests more in-depth data analysis and exploring other sociodemographic factors, as well as training healthcare software developers in advanced data science. This method may be suitable for other chronic diseases as well.