Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations
Wrappers for feature subset selection
Artificial Intelligence - Special issue on relevance
Toward Integrating Feature Selection Algorithms for Classification and Clustering
IEEE Transactions on Knowledge and Data Engineering
The ANNIGMA-wrapper approach to fast feature selection for neuralnets
IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics
Neural-network feature selector
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
In this paper we propose a new approach to select feature subset based on contribution of input attributes in a three-layered feedforward neural network (NN). Three techniques: constructive, contribution, and backward elimination are integrated together in this method. Initially, to determine the minimal NN architecture, the number of hidden neurons is determined by a constructive approach. After that, one-by-one removal of input attributes is performed to compute their contribution. Finally, a sequential backward elimination is used to generate relevant feature subset from the original input space. The elimination process is continued depending on a criterion. To evaluate the proposed method, we applied it to four real-world benchmark problems. Experimental results confirmed that, the proposed method significantly reduces the irrelevant features without degrading the network performance and generates the feature subset with good generalization ability.