Random Forest Twitter Bot Classifier

As social media usage continues to grow, so does the number of automated bot accounts that either spread malicious content or generate fraudulent popularity for political and social figures. Twitter, one of the more popular social media sites, has been plagued by bot armies and needs to find a way to rid itself of these infestations. Studies have shown that currently there are no accurate ways to detect Twitter bot accounts regularly. Using datasets from IIT (Institute of Informatics and Telematics), a feature set is created that allows a classifier to be both accurate and generalized. This feature set includes accessible features from the Twitter API as well as derivative ratio features that give a different perspective of the account. We then decide upon the Random Forest machine learning algorithm because of its ability to prevent most overfitting as well as create a generalized model that can be deployed for accurate use directly after training. In this paper, we propose a set of attributes for a Random Forest classifier that results in high accuracy (90.25%) and generalizability. To prove our derived feature set outperforms basic feature sets and grants valuable insight, we test our derived features against the most important of the basic features. Our derived ratio features outperform these basic account features.