A ROBUST TECHNIQUE OF FAKE NEWS IDENTIFICATION USING ENSEMBLE FEATURE SELECTION

Abstract

In the modern era, selecting the best feature from the high dimensional dataset with multiple variables has become a challenging task. It has become very prominent to train the model with relevant features eliminating the un-necessary feature. The traditional method feature selection methods are performing their excellence in the field of selecting the feature by creating a subset of feature from the dataset. Though it performs well, sometimes they are not helpful in learning the model by selecting single feature with single classifier. This overfits the model and leads to unnecessary confusion. Therefore, this work implements a Robust Framework of Ensemble feature selection Technique. The ensemble learning combines two or more outputs in which they may be the same type or different types, and they may or may not have been trained on the same training data. This study aims to extract the false news sub features by combining multiple subsets of features using ensemble technique. To get an efficient feature on the fake news opinion pool the Feature Score, Recursive Feature Selection and Elasticnet Feature selection has been used. Finally, feature importance has been created for each feature acts as an aggregator to select the final subset of features. The performance has been analysed with 5 classification algorithms of SVM RF, Logistic Regression, Gradient Boosting Classifier and Ridge classifier. The overall performance with accuracy has been evaluated with each classifier. The best classifier is determined by the highest accuracy rate. Our proposed implemented framework determines that Random Forest acquires a better performance in Accuracy.