Backpropagation network classification using feature interval selection
Feature selection is an important problem in the process of knowledge discovery and classification. Currently there are several methods for feature selection, but most of them try to reduce the feature set by removing feature subsets or by removing full features one at a time in some predetermined order. Sometimes a feature can contain partial information and removal of only selected intervals from that feature would be more appropriate. The goal of this work was to explore the possibilities of feature selection and classification using subintervals and neural networks. The proposed network will learn using subsets of the dataset at a time, and eliminate those that contribute negatively to the classification process. Two approaches for division of attributes are chosen, equal probability and equal interval. The results show that this model can return better accuracy and reduces the data size.