Recommended multiclass classification algortihms for this particular problem

Question

I'm using this dataset, https://archive.ics.uci.edu/ml/datasets/Drug+consumption+%28quantified%29, in a research whose main goal is to find correlations among attributes that influences people to be addicted to some "illegal" drugs. For example, people who are womem, aged between 25-34 and likes too much chocolate, has a probability of x% to also be addicted to cannabis.

After preprocessing de original database, I now have the following attributes:

Inputs:

18-24
25-34
35-44
45-54
55-64
65+
GENDER  
EDUCATION   
N   
E   
O   
A   
C   
I   
SS  
CAFF    
NICOTINE    
CHOC    
ALCOHOL

Classes:

AMPHET
AMYL    
BENZOS  
CANNABIS    
COKE    
CRACK   
ECSTASY 
HEROIN  
KETAMINE    
LEGALH  
LSD 
METH    
MUSH    
VSA

As you can see, this is a multiclass classification problem, where a sample can belong to more than one class at the same time. Below there is a example sample that belongs to several classes:

My question is: which algorithm(s) best suit for this classification and correlation problem? I've already taken a look in Multinomial Naïve Bayes and Decision Trees, but I'm not sure if these algorithms can solve this particular kind of problem well.

OBS. I'm using Python with Scikit-learn.

Thanks in advance.

Answer 1

Naive Bayes doesn't give good results because of the assumption that all the features are independent. Apart from Random Forest, I would suggest to use SVM or MLP from scikit learn. SVM works well when even when the classes are non-linearly separable, but the problem with SVM is that it takes a lot of time to train a model when the dataset is quite large. Apart from this you can use onevsrestclassifier from sklearn.multiclass for better results. http://scikit-learn.org/stable/modules/generated/sklearn.multiclass.OneVsRestClassifier.html#examples-using-sklearn-multiclass-onevsrestclassifier