Opinion Mining (OM), which is also known as Sentiment classification or a Polarity classification, is that binary classification of task labelling of an opinionated text/document that expresses either a positive or an overall opinion that is negative. Several technique of analysis of a subjective information in the classification of sentiment is available. The extraction of feature chooses sufficient features for characterizing the opinion. The feature selection (FS) that is also known as the variable selection, subset selection or attribute reduction available in machine learning. It does select an input of dataset which closely defines its specific outcome. Here in this research a feature selection mechanism that is wrapper based is used based on a heuristic algorithm. The experiments demonstrate that the proposed method improves the efficiency of the classifiers and achieve higher accuracy.