Artificial Neural Networks to Detect Risk of Type 2 Diabetes

Abstract

Type 2 diabetes constitutes 85% - 90% of all cases of diabetes with an expectation of 300 million cases around the world by the year 2025 (Lyssenko et al., 2005). In this research, 7 risk factors and their strength of association to the development of Type 2 diabetes was used as relative weight of input variables. A multilayer feedforward architecture with backpropagation algorithm was designed using Neural Network Toolbox of Matlab. The network was trained using batch mode backpropagation with gradient descent and momentum. Best performed network identified during the training was 2 hidden layers of 6 and 3 neurons, an output layer of 1 neuron, logsigmoid transfer function at the hidden layers and a linear transfer function at the output layer. The network recorded best validation performance of 0.10054 at 552^th epoch and correlation coefficient of 0.99705. A regression plot indicated exact linear relationships with all the axes close to 1. At least 528 out of 1122 of the dataset used were found close to 1, which indicated high risk of Type 2 diabetes.

Keywords: Type 2 diabetes, Risk factors, artificial neural network, Matlab, Backpropagation.