# How to use zero-inflated negative binomial regression for binary classification task?

I am working on a binary classification problem and I am currently employing XGBoost. The dataset consists of several variables which are count variables. The problem is, these features are highly skewed on counts. For example, these are the counts of each value of one of the variables:

0.0     98.175855
1.0      1.275902
2.0      0.348707
3.0      0.199535


I was suggested here to apply Zero-inflated Poisson or Zero-Inflated Negative Binomial regression models. I would like to know how these models work.

1. Are these used for feature transformation which will give me features which I can then feed to XGBoost or should I use them as classifiers?
2. My data is a mixture of count and continuous features. Should I employ these models only on count variables, or all variables.

If someone could suggest a good reference to understand these models intuitively, that would be very helpful.