Multiclass logistic regression with mlogit in R I have a multi-class dataset like the following (a,b,c,d are features and e is the class (it can be 0,1 and 2)).
    a b c d e
1   1 1 2 2 1
2   1 2 4 2 0
3   1 2 4 2 0
4   2 2 2 2 0
5   2 1 2 2 2

I am trying to use mlogit package in order to see which column is more important but I am having a difficulty to understand how to use it.
What I am doing is:
dataset$e<-as.factor(dataset$e)
mldata<-mlogit.data(dataset, choice="e")

> mldata[1:5,]
    a b c d     e chid alt
1.0 1 1 2 2 FALSE    1   0
1.1 1 1 2 2  TRUE    1   1
1.2 1 1 2 2 FALSE    1   2
2.0 1 2 4 2  TRUE    2   0
2.1 1 2 4 2 FALSE    2   1

Now, in order to see the coefficients, I am constructing the model like that: 

mlogit.model<- mlogit(e~1|a+b+c+d, data = mldata, reflevel="1")
  mlogit.model

Call:
mlogit(formula = e ~ 1 | a + b + c + d, data = mldata, reflevel = "1",     method = "nr", print.level = 0)

Coefficients:
     alt0       alt2     alt0:a     alt2:a     alt0:b     alt2:b     alt0:c  
-211.0953   -89.7558    27.9911    33.1440    37.7503     7.3585     8.5072  
   alt2:c     alt0:d     alt2:d  
   3.0950    37.1340     5.5584  

But now I don't understand what are alt0, alt2? What are the real coefficients of a,b,c and d?
 A: Multinomial logit assumes that you have a categorical dependent variable.  In your case, there are three categories, denoted 0, 1, and 2.  You've set 1 as the reference category, which means that mlogit is going to use 1 as the baseline category -- everything else is compared to 1.
The thing to keep in mind is that in the mlogit framework, each additional category is compared against the reference group.  Each has its own log-linear model, with k+1 parameters, where k is the number of predictors in your model, and the 1 accounts for the intercept term.  Since you have 2 additional categories and 4 predictors, you're going to be estimating 2*(4+1) = 10 parameters.
So, alt0 is the intercept term for the 0-against-1 comparison, and alt-a,b,c and d describe the marginal change in likelihood between categories 0 and 1.  The alt2 parameters describe the category 2-against-1 comparison model.
See Train's excellent and (last I checked) free pdf book on discrete choice models for more background.
