A feedforward artificial neural network model for classification and detection of type 2 diabetes

Abstract

Abstract Efforts to enhance accuracy in medical diagnostics in molecular medicine have contributed to the wide use of artificial neural network (ANN) algorithms for disease detection due to its ability to process large medical datasets and integrate them into characterized outputs to avoid misdiagnosis. Typically, the application of ANNs have proven useful in sample analyses of patients with diabetes and in decision support systems. Over the years, various ANN models have been utilized in medical diagnostics; however, these approaches still maintain certain levels of error and have lesser training and testing accuracies in disease detection. In this study, we propose a Feedforward Artificial Neural Network (FFANN) model with a dense neural network architecture suitable for processing numeric and textual dataset. We carefully designed our model structure to have the ability to maximize the number of layers and nodes required to learn every feature of the dataset and also to perform effective computations but avoiding model under fitting and overfitting which occurs when less or more layers are used respectively. This approach puts our model ahead of other state-of-the-art prediction models which have been proposed in terms of performance as it achieved 97.27% and 96.09% training and testing accuracies, respectively, for type 2 diabetes detection on Pima Indian Diabetes dataset.