A Selection Method for Denoising Auto Encoder Features Using Cross Entropy
Autor: | Wei Huang, Jiawei Luo, Sheng Yang, Jie Cai, Shu-Lin Wang |
---|---|
Rok vydání: | 2019 |
Předmět: |
Computer science
Generalization business.industry 0211 other engineering and technologies Pattern recognition Feature selection 02 engineering and technology ComputingMethodologies_PATTERNRECOGNITION Cross entropy Dimension (vector space) 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Noise (video) Artificial intelligence Denoising auto encoder business Selection (genetic algorithm) 021101 geological & geomatics engineering |
Zdroj: | Intelligent Computing Methodologies ISBN: 9783030267650 ICIC (3) |
Popis: | There are a lot of noise and redundant information in gene expression data, which will reduce the accuracy of the classification model. Denoising auto encoder can be used to reduce the dimension for high-dimensional gene expression data, and get high-level features with strong classification ability. In order to get a better classification model furtherly, a high-level feature selection method based on information cross-entropy is proposed. Firstly, denoising auto encoder is used to encode high-dimensional original data to get high-level features. Then, the high-level features with low cross entropy are selected to get the low-dimensional mapping of original data, which is used to generate optimized and simplified classification models. The high-level features obtained by the denoising auto encoder can improve the accuracy of the classification model, and the selection of high-level features can improve the generalization ability of the classification model. The classification accuracy of the new method under different Corruption Level values and selection rate are studied experimentally. Experimental results on several gene expression datasets show that the proposed method is effective. Compared with classical and excellent mRMR and SVM-RFE algorithms furtherly, the proposed method shows better accuracy. |
Databáze: | OpenAIRE |
Externí odkaz: |