Hi, I am doing human activity recognition. In my task, feature scaling gives low accuracy, rather than keeping original feature values. But my feature values are not in either [-1 1] or [0 1]. So, why do I get low accuracy after feature scaling?
Feature scaling usually helps, but it is not guaranteed to improve performance. If you use distance-based methods like SVM, omitting scaling will basically result in models that are disproportionally influenced by the subset of features on a large scale. It may well be the case that those features are in fact the best ones you have. In that case, scaling will reduce performance.
I also faced the similar issue. By the way it actually depends how are you scaling them? Putting all the data together or in pieces. In my experiment, I was dividing the whole data in chunks and then scaling them. When I looked at the values, I realized that feature scaling was ruining the discriminatory. So, I proceeded with not scaling finally.
Even if it doesn't prevent you from fnding good solutions, the search *could* take much longer without scaling.
Here are two nice related discussions. In general, the answer to your questino is imho: "NO,not always, however very often, especially if they are very different." (except if in the possible scaling function we consider the identity function as well :P...) Also, depends what you do with the features, some algorithms may not be affected at all by scaling.
http://stats.stackexchange.com/questions/69157/why-do-we-need-to-normalize-data-before-analysis
https://www.researchgate.net/post/Is_it_necessary_to_normalize_data_before_performing_principle_component_analysis
For Scaling I would say,
Pros:
-Better performance i algorithms like SVMs and Neural networks
-faster conversions
Cons
-you may forget to look for the true coef. in regressions
-won't always guarantee better results
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs