Detection of Fake Twitter Accounts with Multiple Classifier and Data Augmentation Technique

Abstract

Due to the continuous growth of data size on platforms with large data such as social media, the detection of fraudulent accounts on these platforms is becoming more difficult. Although social media is preferred for communication purposes, it is becoming an increasingly attractive target for spammers and fraudsters. A suggestion system can be developed in order to provide better products to customers by analyzing the shares and interactions of people on social media. But if the messages are not sent by a real people, the analysis is wrong. In this study, malicious accounts have been identified in order to prevent dirty and false information circulating on the internet by using the messages sent by social media users. For this purpose, a system has been developed to classify automatic or normal accounts using intelligent techniques. The nearest neighbor, logistic regression and random forest algorithms were used for the identification of counterfeit accounts. The classification performances of these methods were compared and smote and majority voting techniques were applied to related algorithms to improve performance.