Fake account detection in twitter using logistic regression with particle swarm optimization

Online social networks (OSNs) have incredibly grown over the past decades. People connect with their friends, family members, new people and share their views or fun artifact with them through OSNs such as Facebook, Twitter, Instagram, Telegram, etc. However, the global and easy connection have also attracted many fake users to lure the genuine users to fulfill their own malicious intention. In the proposed work, the fake accounts (fake followers) detection in Twitter by selecting the relevant features that define the profile's characteristics users has been studied. Here, logistic regression integrated with the particle swarm optimization (PSO) have been proposed to effectively classify an account as genuine or fake. In addition, opposition-based initialization is combined with PSO to start the exploration of the search space with a good set of solutions. Feature selection techniques, namely information gain, correlation, and minimum redundancy maximum relevance, were also applied to select the informative subset of features from the original feature space. The conducted experiments demonstrate that the proposed model attains better performance compare to the competitive state-of-the-arts in most of the cases.