ANNEbot: An evolutionary artificial neural network framework

摘要

A generic framework that uses evolutionary algorithms to obtain the optimal artificial neural network for a given application is presented. ANNEbot uses the concept of Evolutionary Artificial Neural Networks (EANNs) to search for the optimum network architecture and connection structure that would best suit a given problem, without having to incorporate prior knowledge of the domain into the learning mechanism. The framework was tested on the Iris classification data set and the Parkinson's disease diagnosis data set, both of which provided results with above 95% accuracy. These results were then compared with the results obtained from applying other machine learning algorithms such as back propagation and support vector machines on these two data sets. The comparison showed that ANNEbot provided a higher degree of accuracy in contrast to the more commonly used machine learning algorithms.