Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

I'm dealing with a supervized binary classification issue. My dataset is composed of 1500 individuals, living in 600 households. I have approximately 4000 variables to classify my subjects as "infected/uninfected".

I was wondering how would it be possible to account for the hierarchical nature of my data in a data mining classification method, such as CART, MARS or other methods, as it is done for instance in mixed-effects models ? I suppose that the hierarchical structure of the data cannot be ignored, because the risk of a individual to be infected is higher is there is already an infected individual in his household.

Thank you


share|improve this question
add comment

1 Answer

I don't see the need to be concerned about the household as a factor. It sort of depends on the what these 4000 predictive variables are. I wonder how many of them are actually useful. It might help to narrow that down a little to a more manageable set. Suppose for example you include an indicator variable that is 1 if an individual in your household is infected and zero if not. Then if being in the same household with an infected person greatly raises the odds of you being infected algorithms like CART would pick this up and it would be the first or at least an early splitting variable. Of course that binary variable must be known for predicting with a new person. I guess it depends on how you are going to use this algorithm and you haven't really given us a good idea of what you are doing and why.

share|improve this answer
Hi Michael thank you for your answer. My dataset is composed of predictors concerning biological markers, lifestyle, household characteristics, medical history .... My goal is to build an accurate classifier and to identify which predictors are important and how they modify the risk to be infected or not. That is why I planned to use methods such as CART instead of RF for instance, which I planned to use to select useful predictors. Your idea to create an indicator variable seems to be a good way to deal with my issue. –  yoyo May 25 '12 at 10:47
@yoyo That sounds great. I guess even my conjecturing , guesses and resulting suggestions helped! I am glad for that. What does the acronym RF stand for? –  Michael Chernick May 25 '12 at 13:20
It stands for Random Forest ;) –  yoyo May 25 '12 at 13:37
add comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.