Prediction of Diabetes Using Statistical and Machine Learning Modelling Techniques

Abstract

Statistical and machine learning modelling techniques have been effectively used in the healthcare domain and the prediction of epidemiological chronic diseases such as diabetes, which is classified as an epidemic due to its high rates of global prevalence. These techniques are useful for the processes of description, prediction, and evaluation of various diseases, including diabetes. This paper models diabetes disease in Saudi Arabia using the most relevant risk factors, namely smoking, obesity, and physical inactivity for adults aged ≥25 years. The aim of this study is based on developing statistical and machine learning models for the purpose of studying the trends in incidence rates of diabetes over 15 years (1999–2013) and to obtain predictions for future levels of the disease up to 2025, to support health policy planning and resource allocation for controlling diabetes. Different models were developed, namely Multiple Linear Regression (MLR), Support Vector Regression (SVR), Bayesian Linear Regression (BLM), Adaptive Neuro-Fuzzy Inference model (ANFIS), and Artificial Neural Network (ANN). The performance of the developed models is evaluated using four statistical metrices: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and coefficient of determination R-squared. Based on the results, it can be observed that the overall performance for all proposed models was reasonably good; however, the best results were achieved by the ANFIS model with RMSE = 0.04 and R2 = 0.99 for men’s training data, and RMSE = 0.02 and R2 = 0.99 for women’s training data.