Catalog Home Page

The fuzzy misclassification analysis with deep neural network for handling class noise problem

Nugaliyadde, A., Pruengkarn, R. and Wong, K.W. (2018) The fuzzy misclassification analysis with deep neural network for handling class noise problem. In: International Conference on Neural Information Processing (ICONIP) 2018, 13 - 16 December 2018, Siem Reap, Cambodia

Link to Published Version: https://doi.org/10.1007/978-3-030-04212-7_28
*Subscription may be required

Abstract

Most of the real world data is embedded with noise, and noise can negatively affect the classification learning models which are used to analyse data. Therefore, noisy data should be handled in order to avoid any negative effect on the learning algorithm used to build the analysis model. Deep learning algorithm has shown to outperform general classification algorithms. However, it has undermined by noisy data. This paper proposes a Fuzzy misclassification the analysis with deep neural networks (FAD) to handle the noise in classification ion data. By combining the fuzzy misclassification analysis with the deep neural network, it can improve the classification confidence by better handling the noisy data. The FAD has tested on Ionosphere, Pima, German and Yeast3 datasets by randomly adding 40% of noise to the data. The FAD has shown to consistently provide good results when compared to other noise removal techniques. FAD has outperformed CMTF-SVM by an average of 3.88% in the testing datasets.

Publication Type: Conference Paper
Murdoch Affiliation: School of Engineering and Information Technology
URI: http://researchrepository.murdoch.edu.au/id/eprint/42989
Item Control Page Item Control Page