I have seen literature reporting the use of statistical techniques like Mann-Whitney U test and Cohen's D effect size to identify most suitable subset of features for a classifcation problem. But can't this be achieved by employing feature selection and classification using wrapper method? Is there any advantage in using both i.e. aply statistical techniques first and further employ wrapper method on the derived subset of features?
Dear Anusha a S ,
I think these methods can help you:
Conference Paper Infinite Feature Selection
Conference Paper Infinite Latent Feature Selection: A Probabilistic Latent Gr...
In general, in these papers, the authors propose a novel feature selection algorithm which uses a probabilistic graph-based and considers all the possible subsets of features.
Hi Anusha,
Yes, you're right; the wrapper method can be used then classification (without applying any of the tests you highlighted).
I'd recommend using statistical test(s) in order to compare the performance of several machine learning algorithms when performing feature selection and classification.
HTH.
Dr. Samer Sarsam
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques