A drift aware adaptive method based on minimum uncertainty for anomaly detection in social networking. (30th December 2020)

• Record Type: Journal Article
• Authors: mahmodi, Emad; Yazdi, Hadi Sadoghi; Bafghi, Abbas Ghaemi
• Is Part Of: Expert systems with applications. Volume 162(2020)
• Publication Date: 2020-12-30
• Subjects: Concept drift -- Data stream -- Fusion of experts -- Online learning; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.113881
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The social attack is an example of the anomaly that often changed their behavior, increased data volumes, and should be detected as early as possible to minimize damage. Data streaming mining is one of the solutions, which can handle the social attacks, and adapt to the change in the anomaly data stream. In this paper, we propose Online Fusion of Experts based on a minimum uncertainty to predict the concept drift in a data stream of social network-attack. Online learning algorithms such as Linear-order algorithms and Gaussian-order algorithms employ as an expert to identify the change in the anomaly data stream. First, online learning algorithms determine the error value for each data sample when data stream enter individually. Second, O F E utilizes a maximum-posterior estimation of the error rate of online learning algorithms to generate a new input data stream. Third, the Uncertainty Error Correlation Matrix (UECM) of input data applies to real-time behavior change detection of a data stream. Performance of O F E is evaluated by related data streaming algorithms using a benchmark, and real dataset from UCI repository (NSL-KDD, ISCX, and etc.), and malicious web pages, respectively. Highlights: We present concept drift detection based on online learning algorithms. Our approach provides an adaptive learning system for anomaly detection in data stream. Minimum uncertainty employed to combining online learning algorithm. Anomaly in social network data stream employed as a real world concept drift benchmark.