What impact cause feature scaling on the model performance?

spyKY report abuse

While working with data preprocessing before modelling few questions rose in my head. Is it possible that by applying scaling, some of the significant features will lose their weight, and thus have lower impact to explaining the variance of the response variable? If yes, if some important features would be identified by expert knowledge, is it OK to scale other features but those? Or I need to scale only the significant features?

Answers

srSikh report abuse

Algorithms that work by finding distances or similarities between the data samples ( like K-nearest heighbour and SVM) are quite sensitive to feature modifications. The reason why we apply feature scaling is to make our model robust to outliers and make an initial impact of every feature roughly similar. So, my answer is that you should explore your data more carefully, detect the outliers and apply scaling only if needed.

spyKY report abuse

Thanks for the explanation. But how about expert knowledge? Should I use it?

srSikh report abuse

Not sure it's a good idea.

Add Answer

Need support?

Just drop us an email to ... Show more