Feature Selection using Random Forest Method for Sentiment Analysis

Background/Objectives: Online review has become important decision support system for the customers to decide on the subscription or purchse. This paper is aiming to suggest a method that improves the accuracy of the classifier. Methods/Statistical analysis: Feature selection for sentiment analysis using decision forest method and Principal Component Analysis (PCA) is used for the feature reduction. The proposed method is evaluated using twitter data set. Findings: It is proved, that the proposed decision forest based feature extraction improves the precision of the classifiers in the range of 12.49% to 62.5% when compared to PCA and by 49.5% to 62.5% when compared to decision tree based feature selection. Application/Improvements: This method is applicable to product reviews, emotion detection, Knowledge transformation, and predictive analytics.