Published in Journal of Software Engineering and Applications, Vol. 2014 No. 7 (June 2014) at doi: 10.4236/jsea.2014.77052


The larger the size of the data, structured or unstructured, the harder to understand and make use of it. One of the fundamentals to machine learning is feature selection. Feature selection, by reducing the number of irrelevant/redundant features, dramatically reduces the run time of a learning algorithm and leads to a more general concept. In this paper, realization of feature selection through a neural network based algorithm, with the aid of a topology optimizer genetic algorithm, is investigated. We have utilized NeuroEvolution of Augmenting Topologies (NEAT) to select a subset of features with the most relevant connection to the target concept. Discovery and improvement of solutions are two main goals of machine learning, however, the accuracy of these varies depends on dimensions of problem space. Although feature selection methods can help to improve this accuracy, complexity of problem can also affect their performance. Artificialneural networks are proven effective in feature elimination, but as a consequence of fixed topology of most neural networks, it loses accuracy when the number of local minimas is considerable in the problem. To minimize this drawback, topology of neural network should be flexible and it should be able to avoid local minimas especially when a feature is removed. In this work, the power of feature selection through NEAT method is demonstrated. When compared to the evolution of networks with fixed structure, NEAT discovers significantly more sophisticated strategies. The results show NEAT can provide better accuracy compared to conventional Multi-Layer Perceptron and leads to improved feature selection.