Cascading of RBFN, PNN and SVM for Improved Type-2 Diabetes Prediction Accuracy

Abstract

Diabetes is a metabolic disorder caused by a defect in insulin secretion or action (or both) leading to hyperglycemia (high glucose levels) . Over time, hyperglycemia damages nerves and blood vessels, leading to complications like heart disease, stroke, kidney disease, blindness, nerve problems, gum infections and amputation. In order to increase the classification accuracy on diabetes data in this paper a dual-stage cascaded ensemble framework is proposed. This frame work has two stages, the first stage consists of simple Radial Basis Neural Network (RBFN) and simple Probabilistic Neural Network (PNN). The results from both the neural networks are combined and serve as inputs to the second stage classifier called support vector machine. The soundness of proposed framework is validated using Pima Indians Diabetes dataset. The Experimental results indicate that the proposed Dual stage network out performs individual as well as state of-the-art models.