The Effect of Feature Selection Methods in Detecting Malicious Accounts in Social Media

Abstract

Social network is a significant aspect of our lives since it has become an essential part of global communication. Twitter is a microblogging network platform, with an average of more than half a billion tweets posted per day by millions of users. With such diversity and widespread use, Twitter is easily affected with malicious activities by using fake profiles by hackers to carry out malicious activities, this network is suffering from identity theft via fake accounts. In this work, we propose a method to detect these fake accounts by using the decision tree (DT) - recursive feature elimination (DT-RFE) algorithm as a feature selection method to choose the best features in the personal profiles with data standardization to speed up processing, finally, the efficiency of computational detection was evaluated using a collection of supervised machine learning methods. The study showed a high degree of accuracy in classifying the accounts on all proposed algorithms using the feature selection method, with the highest degree of accuracy of 94% by using the random forest algorithm, and even showed the importance of feature selection methods in enhancing detection accuracy by reducing non-significant features that negatively affect the classification process.