0
$\begingroup$

I am working on a university R project of multivariate analysis and I need some help:

DATA: MIXED, with 17 variables : 4 qualitative and the 13 are continuous. PROBLEM: I don't have a class variable, and I have to create it before being able to do any classification. we have information about the class of of only 50/400 , (class of 2 actors with 50 movies) QUESTION: What would be the best method to create my class variable so that the classification would't be unbalanced by a wrong weight of the created class variable; I have created a binary variable class with value 1 for the 50 movies and value 0 for the 350 others, used a logistic classifier but my accuracy is 0.005!! which made me think that class variable has to be created of considered differently! If the creation of the class variable the way i 've done was correct, is there any other optimal classification model that wouldn't be influenced by the big asymmetry of the class variable? Any suggestion please?

Improve this question
$\endgroup$
2
  • $\begingroup$ I don't really get your problem. You start without class at all or do you know the class for just 50 subjects?. In the first case you might want to look at unsupervised learning methods such as k-means clustering. $\endgroup$
    – PeterD
    Commented Jan 8, 2019 at 19:16
  • $\begingroup$ Thanks Peter,. I don'have a classification variable, but i have to create one, knowing that some 50 of my individuals are classified 'x'. We don't have precise info about the class of the other ones. I tried to create a classification variable in a naive way assigning 1 to the 50 and 0 to the others, but that gives me unbalanced df and bad accuracy results. I also tried to cluster using PAM, and I took the result of my clustering as a classification variable for the classification. I 'm not sure if my approach is correct. Thank you for your feed-back! $\endgroup$
    – C.Camus
    Commented Jan 8, 2019 at 19:25

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.