A Data Science Central Community
You can adopt this strategy by using the properties of PCA and classification.
The interest for these two (2) techniques comes from the fact that the PCA allows by its graphic representation to control the result of the classification obtained either independently or from the principal components. This representation allows to:
1.Easily isolate some individuals whose data appear to be outliers and may come from a measurement or input error ;
2.To select the most unusual individuals of a class, or on the opposite those who are close to a neighbouring class towards which they can move.
I have a powerpoint presentation on this subject if you are interested I can send it to you. it will allow you to better understand the principle