Feature Selection and Transformation
- 2k Downloads
A fundamental problem in pattern classification is to work with a set of features which are appropriate for the classification requirements. The first step is the feature extraction. In image classification, for example, the feature set commonly consists of gradients, salient points, SIFT features, etc. High-level features can also be extracted. For example, the detection of the number of faces and their positions, the detection of walls or surfaces in a structured environment, or text detection are high-level features which also are classification problems in and of themselves.
Once designed the set of features it is convenient to select the most informative of them. The reason for this is that the feature extraction process does not yield the best features for some concrete problems. The original feature set usually contains more features than it is necessary. Some of them could be redundant, and some could introduce noise, or be irrelevant. In some problems the number of features is very high and their dimensionality has to be reduced in order to make the problem tractable. In other problems feature selection provides new knowledge about the data classes. For example, in gene selection  a set of genes (features) are sought in order to explain which genes cause some disease. On the other hand, a properly selected feature set significantly improves classification performance. However, feature selection is a challenging task.
KeywordsFeature Selection Mutual Information Independent Component Analysis Independent Component Analysis Algorithm Feature Selection Process
Unable to display preview. Download preview PDF.
- B. Bonev, F. Escolano, and M. Cazorla. “Feature Selection, Mutual Information, and the Classification of High-Dimensional Patterns”. Pattern Analysis and Applications 1433–7541 (2008)Google Scholar
- N. Vasconcelos and M. Vasconcelos. “Scalable Discriminant Feature Selection for Image Retrieval and Recognition”. Computer Vision and Pattern Recognition Conference, Washington, DC (USA) (2004)Google Scholar
- D. Koller and M. Sahami. “Toward Optimal Feature Selection”. ICML-96: Proceedings of the Thirteenth International Conference on Machine Learning, pp. 284–292, San Francisco, CA: Morgan Kaufmann, Bari (Italy) (1996)Google Scholar