Application of modeling techniques to diabetes diagnosis

Abstract

In recent times, the introduction of complex-valued neural networks (CVNN) has widened the scope and applications of real-valued neural network (RVNN) and parametric modeling techniques. In this paper, new expert systems for automatic diagnosis and classification of diabetes using CVNN and RVNN based parametric modeling approaches have been suggested. Application of complex data normalization technique converts the real valued input data to complex valued data (CVD) by the process of phase encoding over unity magnitude. CVNN learn the relationship between the input and output phase encoded data during training and the coefficients of Complex-valued autoregressive (CAR) model can be extracted from the complex-valued weights and coefficients of the trained network. Classification of the obtained CAR or RVAR model coefficients results in required distinct classes for diagnosis purpose. Similar operations can be performed for real-valued autoregressive technique except for CVD normalization. The effect of data normalization techniques, activation functions, learning rate, number of neurons in the hidden layer and the number of epoch using the suggested techniques on PIMA INDIA diabetes dataset have been evaluated in this paper. Results obtained compares favorably with earlier reported results.