Detecting Fake Accounts in Social Networks Using Principal ComponentsAnalysis and KernelDensity Estimation Algorithm (A Case Study on the Twitter Social Network)

The use of social networks is growing increasingly and people spend a lot of their time using thesenetworks. Celebrities and companies have used these networks to connect with their fans and customers andnews agencies use these networks to publish news. In line with the growing popularity of online socialnetworks, security risks and threats are also increasing, and malicious activities and attacks such asphishing, creating fake accounts and spam on these networks have increased significantly. In a fake accountattack, malicious users introduce themselves instead of other people by creating a fake account and in thisway, they abuse the reputation of individuals or companies. This paper presents a new method for detectingfake accounts in social networks based on machine learning algorithms. The proposed method for machinetraining uses Various similarity features such as Cosine similarity, Jaccard similarity, friendship networksimilarity, and centrality measures. All these features are extracted from the graph adjacency matrix of thesocial network. Then, principal component analysis was used in order to reduce the data dimensions andsolve the problem of overfitting. The data are then classified using the Kernel Density Estimationclassification and the Self Organization map and the results of the proposed method are evaluated using themeasure of accuracy, sensitivity, and false-positive rate. Examination of the results shows that the proposedmethod detects fake accounts with 99.6% accuracy which is about 5% better than Cao's method. The rate ofmisdiagnosis of fake accounts also improved by 3% compared to the same method.