Study of Type 2 diabetes risk factors using neural network for Thai people and tuning neural network parameters

Risk factors for Type 2 diabetes is very important for developing diabetes prediction tools instead of blood testing. Recently, many researches have studied risk factors of diabetes in order to apply them to be a tool for diabetes prediction by using Logistic Regression (LR), Radial Basis and Back-propagation Neural Network (BNN). However, the accuracy is not higher. This paper presents new factors that are smoking and alcohol consumption to improve accuracy in diabetes prediction. Some traditional factors i.e., body mass index (BMI), blood pressure (BP) and waist circumference (WC) and Family History (FMH) are also proposed to extent by adjusting and additional range. The proposed diabetes prediction method is based on BNN. Approximately 2,000 cases of Thai people at BMC hospital, Thailand during 2010 to 2012 are used to train the BNN. From experiment results, each proposed factors i.e., FMH, Alcohol consumption factor, Smoking Factors and WC gives a value of accuracy that is higher than baseline as 83.35%, 83.5%, 83.6% and 83.65%, respectively. After that, this paper focuses on tuning neural network parameter, which is divided into 3 main steps: number of hidden nodes, sequence of integrating the proposed factors, and other parameter i.e., learning rate, and Iteration. Finally, the proposed factors and tuning BNN parameters introduce a high accuracy compared with the baseline up to 1.2%.