Whatever features we give are just numerical/categorical values for a machine learning algorithm. It will just try to find out a pattern in it. Different feature selection techniques return different features to be important. Most of the comparison studies fix the classifier so as to compare different feature selection technique performances in improving classification by the classifier. But is it possible to compare feature selection technique performance based on its relevance for explaining the training dataset? The relevance can either be qualitative, domain knowledge based or based on a logical comparison metric. Are there any studies that performs such comparison?
There must be some research paper in this filed. Mostly these feature selection depends on the dataset. Not always, I think, same types of result will come in all scenarios except some statistical features.
There is another thing called manifold learning. In these techniques, we can reduce the dimension of the data. So it means, these reduced data hold some identical information as well. So if you look for the differences between different manifold learning technique, I think you will get some direct resources without specify any field.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training