FGANN: A Hybrid Approach for Medical Diagnosing

Abstract

Medical diagnostic support systems often deal with a large number of disease measurements and relatively small number of patient records. All these measurements (features) may not be relevant for diagnosing, and some may contain noise due to human or machine errors. These features greatly affect the results of diagnostic systems, and this is significantly high with less number of available patient records. Further, these features will guzzle memory space and time required for diagnosis process. These issues have been addressed in the proposed approach FGANN, which is fuzzy genetic algorithm-based neural network for prediction of disease outcome in the field of health care. In this proposed hybrid approach, the feature space has been modeled using fuzzy approach and then genetic algorithm (GA) has been employed to extract prominent features that show vital impact on diagnosis. These obtained key features are used to train neural network (NN) which in turn used to predict the outcome of the disease for a given patient record. The experiments were carried out on two different types of diseases like diabetics and thyroid by considering standard datasets. In this hybrid approach, not only prediction accuracy, but also the time taken by NN for learning and memory space occupied by patient’s information that has been considered as performance measures of the system. The results showed that proposed approach fuzzy logic + GA + NN giving more accurate measures of diagnosis compared to an approach based on NN.