An accurate attack detection framework based on exponential polynomial kernel‐centered deep neural networks in the wireless sensor network

Abstract

AbstractA novel type of wireless network that is rising in popularity with a wide assortment of civilian along with military applications is called WSN. Owing to several factors like nodes with resources that are constrained and packages that are resistant to tamper, WSNs are tremendously vulnerable to internal attacks and mainly external attacks also. Attacks on such critical schemes comprise penetrations into their network along with the installation of malicious tools or programs, which can reveal sensitive data or alter the specific physical equipment's behavior. For their action, the wireless networks are utilized by the threats like spoofing, injection, denial of services, and numerous attacks. Thus, for protecting devices from intruder attacks, security solutions are necessary. An intrusion detection system (IDS), which is wielded for detecting attacks against a system or a network by evaluating their activities along with events, is a tool. In this article, an efficient attack detection technique grounded on exponential polynomial kernel‐centered deep neural networks (EPK‐DNN) is proposed since intrusion detection is crucial in securing the data. Intrusion detection in WSN is extremely intricate for tasks like fault diagnosis, and real‐time monitoring applications, owing to the WSN's dynamicity. To find diverse attacks along with to safeguard WSNs from security risks, numerous detection methodologies are created, because intrusion detection is decisive for protecting the data in WSNs. However, owing to the restricted resources and energy of WSN nodes, widespread computation and so forth, they are inefficient. In this article, an efficient attack detection methodology centered on EPK‐DNN is proposed to lessen these problems. The attack detection system's training is the foremost step in the EPK‐DNN technique. In step one, the input data are preprocessed; and then, in the training process, the preprocessed data are exploited for attribute extraction. In step two, by utilizing the linear scaling based BAT optimization (LS‐BAT), the major attributes are chosen. Then, to detect attacks in WSN, the chosen features are trained by the EPK‐DNN. In step three, by utilizing the Damerau‐Levenshtein‐based K‐means algorithm (DL‐K‐Means), the WSN network is initialized along with the sensor nodes are clustered. To amass the sensor data, the cluster heads are selected by utilizing the swap, displacement, and reversion‐centered rock hyraxes swarm optimization algorithm. After that, for testing, the sensor data are inputted into the trained ADS. The outcomes exhibited that the greatest accuracy rate of 97.21% was attained by the EPK‐DNN technique for the real‐time BC dataset and 96.86% for the real‐time MC dataset. When analogized to the customary deep learning (DL) methodologies, the investigational findings reveal that the EPK‐DNN technique accomplishes desirable detection accuracy.