Fake news detection using ensemble machine learning

Consuming news from social networks has become the new normal. In theory, the rapid rise of social media can be seen to have a positive impact in promoting active citizenship. Unfortunately, social networks can also prove to be a considerable threat. Just as their platforms promote easy access to rapid and low-cost dissemination of information, social media is also fertile ground for the spreading of misinformation. As a result, governments now face the reality of having to bolt the proverbial barn door while the fake news horse is already free, running amok. Given fake news’ ability to deceive, cause instability, and spread propaganda, governments must ensure that there are measures in place that will allow them to effectively deal with what qualifies as a form of cyber warfare. It is therefore quite critical to be able to detect fake news on social media to mitigate the potential negative effects. Detecting fake news on social networks can be quite a challenge as they are often written to masquerade as real news. To effectively detect fake news on social networks, one may need to exhaust all the auxiliary information. Furthermore, the sheer amount at which fake news is seen to propagate means that the process of identifying and shutting down the fake news articles cannot be left to human means alone. Ensemble machine learning makes use of a set of classifiers whose individual decisions are aggregated by weighted voting to improve predictions, decrease variance and bias. In this paper, we present an ensemble machine learning model which determines the truth probability of given statements from social networks by considering all the relating metadata. The proposed system makes use of five different classifiers to improve the detection of fake news on social networks. We trained and tested the model using a fake news dataset with the results of our experiment yielding fake news detection accuracies averaging 80%. Results are shown to improve significantly when the feature selection stage of the training process includes more attributes of the dataset.