An Improvement of Empirical Risk Functional in Neuro-Fuzzy Classifier

This paper suggests a new method to improve of Empirical Risk Functional . Empirical Risk Functional acts as cost function for training neuro-fuzzy classifiers. Empirical risk minimization seeks the function that best fits the training data and it is equivalent to maximum likelihood estimation. The name of this cost function is Approximate Differentiable Empirical Risk Functional (ADERF).This function enables us to use a differentiable approximation of the misclassification rate so that the Empirical Risk Minimization Principle formulated in Statistical Learning Theory can be applied. Also there is a learning algorithm based on ADERF. With our new method,more component of output vector of fuzzy classifier map to 1.By evaluating the effects of the proposed method, we can see the convergence speed of the learning algorithm and the classification accuracy are improved,and causes improved ADERF. The effects of improved ADERF, was illustrated. Experimental results on a number of benchmark classification tasks and comparison between approaches are provided.