NeuroEvolutionary Feature Selection Using NEATReportar como inadecuado

NeuroEvolutionary Feature Selection Using NEAT - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

The larger the size of thedata, structured or unstructured, the harder to understand and make use of it.One of the fundamentals to machine learning is feature selection. Featureselection, by reducing the number of irrelevant-redundant features,dramatically reduces the run time of a learning algorithm and leads to a moregeneral concept. In this paper, realization of feature selection through aneural network based algorithm, with the aid of a topology optimizer geneticalgorithm, is investigated. We have utilized NeuroEvolution of AugmentingTopologies NEAT to select a subset of features with the most relevantconnection to the target concept. Discovery and improvement of solutions aretwo main goals of machine learning, however, the accuracy of these variesdepends on dimensions of problem space. Although feature selection methods canhelp to improve this accuracy, complexity of problem can also affect theirperformance. Artificialneural networks are proven effective in featureelimination, but as a consequence of fixed topology of most neural networks, itloses accuracy when the number of local minimas is considerable in the problem.To minimize this drawback, topology of neural network should be flexible and itshould be able to avoid local minimas especially when a feature is removed. Inthis work, the power of feature selection through NEAT method is demonstrated.When compared to the evolution of networks with fixed structure, NEAT discoverssignificantly more sophisticated strategies. The results show NEAT can providebetter accuracy compared to conventional Multi-Layer Perceptron and leads toimproved feature selection.


NeuroEvolutionary, Feature Selection, NEAT

Cite this paper

Sohangir, S. , Rahimi, S. and Gupta, B. 2014 NeuroEvolutionary Feature Selection Using NEAT. Journal of Software Engineering and Applications, 7, 562-570. doi: 10.4236-jsea.2014.77052.

Autor: Soroosh Sohangir, Shahram Rahimi, Bidyut Gupta



Documentos relacionados