A type-2 diabetes early warning system using particle swarm optimized artificial neural networks: a case of trans Nzoia county of Kenya

Abstract

This research aimed at developing an early warning system for pre-diabetic and diabetics by analyzing simple and easily determinable signs and symptoms of diabetes among the people living in Trans Nzoia County of Kenya using Particle Swarm Optimized Artificial Neural Networks. With the skyrocketing prevalence of type 2 diabetes in Kenya the system can be used to encourage affected people to seek further medical attention to prevent the onset of diabetes or start managing it early enough to avoid the associated complications. The study sought to find out the best predictive variables of Type 2 Diabetes Mellitus, developed a system to diagnose diabetes from the variables using Artificial Neural Networks and tested the system on accuracy to find out the effectiveness of the system as an early warning system for the disease. Data was collected from diabetes clinics in hospitals within Trans Nzoia County of Kenya. The collected data was first preprocessed by R software to select the best generalizing attributes which were thereafter used to model an Artificial Neural Network. Particle Swarm Optimization algorithm was used to explore the global minima of the solution curve which was later exploited using the back propagation algorithm. The network attained a 70% and 66.23% accuracy on training and test data respectively. The network also attained a sensitivity of 70.23% and a specificity of 64.24%. This clearly shows that the system can be used as an early warning system for type 2 diabetes mellitus.