Developing Transformed Fuzzy Neural Networks to Enhance Medical Data Classification Accuracy

Developing an automatic model to precisely and quickly classify medical data is a challenging work due to lack of medical knowledge. This paper presented a transformed fuzzy neural network (TFNN) to enhance medical data classification accuracy by learning knowledge from the available medical databases. We developed a novel reward/penalty fuzzy rule base to infer the degree of transformation for each input attribute. The degree of reward/penalty was determined according to its original measured value relative to its attribute mean and standard deviation. The farther the deviation of a datum to its mean value, the larger the penalty was exerted on the datum. In contrast, a datum in the neighborhood of mean value received a positive award. Furthermore, a pattern’s desired output was considered a valuable knowledge and was treated as an extra input to the TFNN. In this study, three medical datasets are used to validate the classification accuracy. Results compared to the back propagation neural network and support vector machine verified the effectiveness of the proposed model in medical data classification.