@mistry , Hope this answer will help you understand better.

** Discriminative Approach**- This happens in the case of logistic regression, Conditional Random Fields (CRFs), etc. Some literature also describes these as probabilistic classifiers. The name ‘discriminative’ comes from the fact that under a probabilistic perspective, we are trying to find a conditional distribution that is helping us differentiate between different labels. In this approach, we don’t care about the distribution of data.

**Generative Approaches** - When we try to predict Pr(x|y) or more generally Pr(x,y) from the data ,we are trying to model the distribution D from which the data is generated.

**For example**-

Let’s try to predict Pr(y|x)

Say (x,y) stand for (Height,Gender) in a country

A discriminative approach tells us that for a given height, how much the height is likely to be of a female or a male. The generative approach will let us know the distribution of heights for each female and a male.

The discriminative approach helps us make predictions based on gender from a country while the generative approach can help us make comparisons between different countries in terms of their gender heights.