Abstract:
The penetration of social and online platforms has opened a new substantial domain of Fake news dissemination in the current time. Also, this dynamic form of data opens up new dimensions for researchers to detect Fake news from the ocean of data. Therefore, Fake news detection has attracted both academia and industry indifferently as research or analytical domain in the concurrent time. Due to data availability, the classification tasks have been tested in different sets and types of data. Detecting Fake news evolves as an actual potential domain to explore with more efficient algorithms and parameter-based modified algorithms. In this work, an analytical sketch has been drawn to compare the performances of different classifiers depending on accuracy and time. Seven classifiers of four different types have been implemented and tested namely, Multilayer Perceptron, Sequential Minimal Optimization, Logistic Regression, Decision Tree, J48, Random Forest and Naïve Bayes Classifier. The analytical evaluation process has been designed with three experimental setups, 10-fold cross-validation, 70% split and 80% split. The separate setups show distinctive outcomes across the algorithms. Naïve-Bayes classifier model shows its prominence along with the Random Forest classifier. However, the and Decision Tree-based classifiers perform differently from earlier knowledge. Furthermore, this paper identifies a different aspect of using testing-training splitting in classifier tasks.
