Title: A Neural Network Approach for Attribute Significance Estimation

Year of Publication: Sep - 2017
Page Numbers: 21-25
Authors: Peter Italo De Battista, Alexander Buhmann, Yiannos Manoli
Conference Name: The Fourth International Conference on Artificial Intelligence and Pattern Recognition (AIPR2017)
- Poland


Attribute selection methods explore the interrelationship between the data to avoid less relevant attributes. Some selector methods are also able to estimate the significance rate of input attributes. Removing unnecessary input attributes has several advantages, like a lower variance and complexity of the machine learning model. In this paper, we propose a four-layer feedforward neural network, which estimates the input attribute relevance rate depending on the desired output. The neural network contains a pre-input layer, where every input attribute is connected by a salient weight to the next layer. Therefore, every attribute primarily depends on its salient weight. Two penalty terms are added related to the salient weights. Thus, the relevant and irrelevant attributes can be distinguished. The attribute significance estimation capability of the proposed neural network was evaluated for three artificial and one real regression in addition to a real classification problem.