Swarm BIAS-variance analysis of an evolutionary neural network classifier

In the search of a system satisfying all the requirements of an automated classifier, evolutionary artificial neural networks have been successfully applied to a number of problems, in particular to problems where there is not a deep knowledge of the phenomena and other methods tend to fail. There are many neural models that efficiently solve either function approximation problems in general terms or some particular problems like classification, pattern recognition, clustering and time-series prediction. This success is due to these models main characteristics, in particular those matching the essentials for an automated classifier: keeping bias and variance low. This paper presents a classifier system based on the benefits arising from the interaction between evolutionary algorithms, such as particle swarm optimization, and artificial neural networks. And a bias variance decomposition of the predictive error shows that the success of the proposed approach lies in the ability of the learning algorithm to properly tune the bias/variance trade-off to reduce the prediction error. To measure the performance, the purposed classifier will be tested on three different well known benchmark problems: the Fisher Iris data set, the Australian credit card assessment and the Pima diabetes data set.