Authors: Bhanu Chander; Kumaravelan Gopalakrishnan
Addresses: Department of Computer Science and Engineering, Pondicherry University, Pondicherry, 609605, India ' Department of Computer Science and Engineering, Pondicherry University, Pondicherry, 609605, India
Abstract: Due to limited resources and harsh deployment environments, data outliers frequently rise in wireless sensor networks (WSNs). Hence, the collected data observations contain poor data quality and reliability. In recent years, research attempts have focused on utilising temporal and spatial correlation of the sensed data in WSNs but ignored the dependencies among the sensor node's attributes, which reduce overall communication. Instead of transmitting every sensed data of a corresponding sensor node to the base station, this paper pursues a novel approach to incorporating a representation method using an auto-encoder to identify the redundant data in its transmission path through cluster head (CH). With this scenario, this paper also empirically assesses the integration of auto-encoders and SVDD to learn a condensed form of a low dimensional data point by interpolating the convex combination of the sensed data, which can semantically mix their characteristics in a distributed manner and identify the outlier respectively.
Keywords: wireless sensor network; WSN; anomaly detection; outlier; auto-encoder; support vector data description; Parzen neural networks.
International Journal of Information and Computer Security, 2023 Vol.22 No.1, pp.28 - 59
Received: 23 Aug 2021
Accepted: 06 Feb 2022
Published online: 14 Sep 2023 *