Title: Auto-encoder-based technique for effective detection of frauds in social networks
Authors: S. Jamuna Rani; S. Vagdevi
Addresses: Department of Computer Science, Visveswaraya Technological University, Karnataka, India ' Department of Computer Science, Visveswaraya Technological University, Karnataka, India
Abstract: Detection of these spam accounts has recently attracted significant attraction in the literature. Most of the spam-account detection techniques presented in the literature employ supervised learning models to achieve their goal. These models require sufficient size of spam-account samples in their training set to be trained effectively. However, obtaining such large sample sizes is a significant challenge. In many real-world scenarios, the number of such available samples is extremely limited. Due to this limitation in the training set, the spam-account detection techniques can exhibit extremely poor detection accuracy. Hence, in this paper, an effective supervised learning model-based spam-account detection technique is presented, which utilises only limited size of spam-account samples in its training set, and to achieve this desired goal, the dimension of the feature vectors in the training set is reduced through the aid of auto-encoders. Further, the spam-accounts are detected based on their corresponding hazard rates. The hazard rates are generated through recurrent neural network. An empirical analysis study is presented, in which, the proposed spam-account detection technique is compared against the contemporary technique. In this study, the proposed technique exhibits relatively superior performance in-terms of classification accuracy.
Keywords: social networks; survival analysis model; fraud detection.
International Journal of Information and Computer Security, 2022 Vol.18 No.3/4, pp.348 - 364
Received: 04 Mar 2021
Accepted: 23 Jul 2021
Published online: 05 Sep 2022 *