Should I concern about multicollinearity when building KNN model for the classification problem?
If yes. how to deal with it?
Asked
Active
Viewed 2,900 times
5
-
This would help, have a look : https://stats.stackexchange.com/a/8284/120788 – jeza Jul 31 '18 at 02:08
1 Answers
5
Clustering analysis is not negatively affected by heteroscedasticity but the results are negatively impacted by multicollinearity of features/ variables used in clustering as the correlated feature/ variable will carry extra weight on the distance calculation than desired.
Solution: Do not drop any variables, but you can use principal component analysis and consider principal component analysis results.
sycamore24
- 53
- 8
-
1do you mean ”classification or cluster”. Also, should I apply PCA first and then to build KNN? – Peter Smith Jul 31 '18 at 20:50
-
The application of PCA first is a subject already in the literature. You are right. – sycamore24 Aug 01 '18 at 06:42
-
1
-
@sycamore24, I did what have said. PCA then KNN. It becomes faster than before but the result is not better than KNN alone. – jeza Aug 14 '18 at 16:20
-