Using feature selection techniques to produce smaller neural networks with better generalisation capabilities

The issue of feature selection is of considerable importance, particularly where artificial neural networks are used, as the size of the network is directly related to the number of input sources. Despite the fact that artificial neural networks have been applied to solve many problems in different fields, and found to be superior to conventional statistical classifiers, they have a major drawback: the need to define the optimum network size for a particular problem. In remote sensing applications, which are generally in the area of image classification, the use of more input features would make the network overspecific to the training data. Over-specificity reduces the generalisation capabilities of a neural network.