Feature Selection is the actual part where the text analytics take place. So how does this happen. For this let us go back a few steps. We started off by saying that we have a data of 1.6 million tweets each having a declared sentiment of good or bad. In this case we would try to understand the predictive power of a feature i.e. a word. For example the word traffic has a high predictive power means that whenever the word is used there is a higher possibility of the sentence being of a particular sentiment.