Design of a predictor for COVID-19 misinformation prediction

ABSTRACT

Due to the increase of social media usage, the online sharing of content has been extremely increased. As a result, the spread of misinformation on social media platforms has also increased. To address this issue, we proposed an approach that predicts the news is fake or real. In our approach, we select the top k ranked features through a filter base algorithm and feed them to the classifier. The main objective of this research is to provide two things. First, to provide an approach, which compares the benchmark performance results of the evolutionary detection approach on the Koirala dataset. The second is to build, publicly available dataset through web scraping for the classification of COVID-19 fake news articles. Our method significantly uplifts the F1-score with 14.88 percent for the same number of features selected 605 for the already existing approach. Also, stated the number of features 5000 on which the approach showed the best results with a margin of F1-score of 20.4 percent, respectively. Similarly, on the self-build dataset, this approach also outshines and achieved 99.66 percent of F1-score, respectively. Our experimental results show that our robust approach by comparing with other classifiers and existing approach, Max-Min Ratio (MMR) along with support vector machine (SVM) outperformed on both of these datasets. Hence, feature selection plays a vital role in the performance of the model rather than deeply tuning and training the classifier.