A practical approach to feature selection
ML92 Proceedings of the ninth international workshop on Machine learning
Estimating attributes: analysis and extensions of RELIEF
ECML-94 Proceedings of the European conference on machine learning on Machine Learning
An adaptation of Relief for attribute estimation in regression
ICML '97 Proceedings of the Fourteenth International Conference on Machine Learning
Inference for the Generalization Error
Machine Learning
Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)
Statistical Comparisons of Classifiers over Multiple Data Sets
The Journal of Machine Learning Research
A review of feature selection techniques in bioinformatics
Bioinformatics
An experimental comparison of performance measures for classification
Pattern Recognition Letters
Feature Selection and Classification for Small Gene Sets
PRIB '08 Proceedings of the Third IAPR International Conference on Pattern Recognition in Bioinformatics
Feature words that classify problem sentence in scientific article
Proceedings of the 14th International Conference on Information Integration and Web-based Applications & Services
Selection of interdependent genes via dynamic relevance analysis for cancer diagnosis
Journal of Biomedical Informatics
Hi-index | 0.00 |
Due to the high number of gene expressions contained on microarray data, feature extraction techniques are usually applied before inducing classifiers. A common criterion to decide on the number of selected genes is minimizing the classifier error. However, considering the risk of overfitting due to the small sample size, and the fact that the number of selected genes is usually larger than the suspected number of discriminating genes, this work proposes relaxing the minimum error rate criterion. The paper shows that from a small number of feature selection and classification methods, it is possible to find configurations that select few genes without significantly worsening the error rate of the best classifier. Average ranking for 10 to 40 genes shows that SVM-RFE with Naïve Bayes and FCBF with SVM behave consistently well.