Anomaly activity detection framework using artificial immune system approach in sensor network

Abstract

Wireless sensor network (WSN) is well known as a type of wireless ad hoc networks. It consists of a numbers of small scale nodes that operating autonomously to perform specific tasks. Due to the limitation of the computational power, sensor nodes are deployed without a centralized authority that would control the flow of individual data traffic. In this dissertation, an anomaly activity detection framework has been proposed for wireless sensor network. The proposed framework is based on the concept of artificial immune system which capable of learning the structure of self body (normal) and detects potentially harmful foreign antigens (abnormal) in a non-biological environment. A Naïve Bayes classifier is introduced in extract sensor network traffic activity to construct the threshold metrics for anomaly detection. Meanwhile, a self tolerant process is used to upgrade the anomaly detection threshold metric every time it encounters a sensor network activity to improve the detection rate (accuracy) and to reduce the false positive rate. The proposed anomaly activity detection framework developed in Dynamic C and embedded into RabbitCore RCM4500W microprocessor. To simulate a wireless sensor network environment, the wireless sensor network collaborates with ZigBee. In the experiment, a wireless sensor network consists of six sensor nodes has been established in an ad hoc architecture. The Denial of Service (DoS) attack has been chosen as the source of attack to evaluate the effectiveness of the proposed framework. As the result, the proposed anomaly activity detection framework showed a 72% detection rate in anomaly activity detection of wireless sensor network traffic activity at the end of the experiment.