FAST CORRELATION DEEP BELIEF AND RAPHSON GRADIENT BOOSTING ENSEMBLE CLASSIFIER FOR DOS ATTACK DETECTION IN WSN

Abstract

In today's contemporary world, the utilization of technology is inevitable and the swift
advancements in the Internet and communication fields have emerged to diversify the Wireless
Sensor Network (WSN) technologies. Enormous devices collect several sensory data for a wide
range of fields and applications. However, WSN has been evinced to be susceptible to security
lapses, integrated with their limited resources and voluminous of data generated instigate a
crucial security concern. In this context, the objective remains in designing significant Denial of
Service (DoS) attack detection by applying a salient machine learning abstraction known as
ensemble learning in order to improve detection performance. The proposed method is called,
Fast Correlation Deep Belief and Raphson Gradient Ensemble Classifier (FCDB-RGEC) for DoS
attack detection in WSN. However, most accessible datasets consist of multiclass output data
with instable distributions, that remain to be the major pitfalls for attack detection accuracy
reduction. Therefore, first Min-Max Normalization-based Preprocessing algorithm is designed
with the normalization function that fix as instabilities identified in the raw dataset. Second, with
the normalized network samples as input, pertinent features are extracted by means of Fast
Correlation-based Deep Belief Network Feature Extraction algorithm. Finally, with the extracted
features, by applying the Raphson Gradient Boosting Ensemble Classifier algorithm, the detection
and classification of four kinds of DoS attacks in WSN have been detected. Moreover, the WSN-DS
The dataset was utilized to examine the efficiency of FCDB-RGEC. Results illustrate significant
improvement in attack detection time, false alarm, recall, as well as precision, as compared to
existing methods.