PCA fusion for ANN-based diabetes diagnostic

Abstract

Diabetes is a result of inability to respond felicitously to insulin requirement for glucose regulation (sugar). In this paper, diabetes detection system is developed utilizing Principal Component Analysis (PCA) and Multilayer Perceptron Artificial Neural Network (MLPANN). Primary investigation focuses on combining source information and PCA transformed features under MLPANNframework. Confusion matrix based analysis has been performed to analysis the effect of source and PCA information fusion. In analysis standard UCI diabetes dataset, the maximum accuracy of 76.5% has been recorder for source features and accuracy of 85.2% with 6 level PCA features while fusion resulted in highest success rate of 87.8%. It acquires 15% and 3% relative accuracy increase when compared with source and PCA features used alone, respectively.