Simultaneous evolution of feature subset and neural classifier on high-dimensional data
Jennifer Hallinan, Paul Jackway · 1999
This paper describes a novel feature selection algorithm which utilises a genetic algorithm to simultaneously optimise a feature subset and the weights for a three-layer feedforward neural network classifier. On the "sonar" data set from UC Irvine, this approach produces results comparable to those reported for other algorithms on the same data, but using fewer input features and a simpler neural network architecture. These results indicate that tailoring a neural network classifier to a specific subset of features has the potential to produce a classifier with low classification error, good generalisability, and relatively low computational overhead. 1. Introduction Many real world problems admit a large number of possible measurements on the data, and therefore by forming products, ratios, moments, transforms, and non-linear functions of these basic measurements, an even larger number of potential features can be computed for the purposes of classification. Usually, we proceed on t...