Fake News Detection - A Comparative Study of Advanced Ensemble Approaches

摘要

People have taken to social media as a platform for gathering news for updates on everything such as entertainment, politics, government, sports, education, and technology. Therefore, it is highly necessary to ensure that the information accessible to the public is reliable. But there is a high risk of malicious forces spreading wrong information, making even the savviest news audience at risk. Misinformation can seriously, indeed fatally harm the community. Many machine learning algorithms were proposed for identifying false news. In this paper, we have analyzed the accuracies between the ensemble model (LSVM, Naive Bayes, Decision Tree) using Voting Classifier with Bagging meta-estimator and Boosting (Adaboost and Gradient Boosting) Classifiers. We make a comparison between the two most effective advanced ensemble algorithms bagging and boosting. We base this analysis on two datasets where one is of a smaller size and another is a bigger dataset. Voting Classifier is used for ensembling LSVM, Naive Bayes, and Decision Tree for both the datasets. This ensembled model has been taken as the base estimator of Bagging meta-estimator and is compared with the other classifiers such as Bagging with default base estimator, Adaboost and Gradient Boosting. Our study shows that Bagging meta-estimator with ensemble model as its base estimator shows better performance than Boosting classifiers for both the datasets.