期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2019
卷号：97
期号：7
页码：1937-1947
出版社：Journal of Theoretical and Applied
摘要：Facebook is on constant growing, attracting more users due to the provided high-quality services for Online Socializing, Sharing Information, Communication, and alike. Facebook manages data for billions of people and is therefore be a target for attacking. As a result, sophisticated ways for infiltrating and threatening this platform have been developed. Fake profiles, for instance, are created for malicious purposes such as financial fraud, impersonate identities, Spamming, etc. Numerous studies have investigated the possibility of detecting fake profiles on Facebook with each study focusing on introducing a new set of features and employing different machine learning algorithms for countermeasure. This paper adopts a set of features from the previous studies and introduces additional features to improve classification performance in order to detect fake profiles. The performance of five supervised algorithms (Decision Tree, Support Vector Machine SVM, Na�ve Bayes, Random Forrest, and k Nearest Neighbour k-NN) are evaluated across three of the common mining tools (RapidMiner, WEKA, and Orange). The experimental results showed that SVM, Na�ve Bayes, and Random Forest had a stable performance with a nearly identical results across the three mining tools. However, Decision Tree outperformed other classifiers on RapidMiner and WEKA with accuracy of 0.9888 and 0.9827, respectively. Finally, we observed that k-NN showed the most significant change with an accuracy of 0.9603 for WEKA, 0.9145 for Orange, and 0.9460 for RapidMiner tool. These findings would be useful for researchers willing to develop a machine learning model to detect malicious activities on social network.
