A study on effective feature extraction and genetic algorithm based feature selection method in fake news detection classification using machine learning approaches

Abstract

In today's technology, information spreads quickly through online social networks, making our lives easier. However, when false news is shared without critical evaluation, it can harm society and affect social, political and economic aspects as it reaches a wide audience. At this point, it is important to develop content verification and confirmation systems. In this study, the aim is to conduct monolingual and cross-lingual classification on a multi-class dataset containing English and German news content. We applied data preprocessing, including CountVectorizer and stylometric feature extraction, before classification. Feature selection was made using the genetic algorithm, which is an algorithm based on the idea of evolution in nature. Selected features were classified by Random Forest, Logistic Regression, Multinomial Naive Bayes, Decision Tree and KNearest Neighbors machine learning algorithms. In the classification process, Multinomial Naive Bayes achieved 58.49% Accuracy and 42.97% macro-F1 for monolingual English news texts, while Logistic Regression achieved 45.39% Accuracy and 37.70% macro-F1 in Cross-lingual classification using English and German news texts. Significantly successful results were obtained compared to studies conducted with the same dataset. In addition, the same methodology was applied to the ISOT dataset. 99.48% and 99.62% macro-F1 were obtained by Logistic Regression and Decision Tree algorithms, respectively.