Multitask Learning-Based Security Event Forecast Methods for Wireless Sensor Networks

Abstract

Wireless sensor networks have strong dynamics and uncertainty, including network topological changes, node disappearance or addition, and facing various threats. First, to strengthen the detection adaptability of wireless sensor networks to various security attacks, a region similarity multitask-based security event forecast method for wireless sensor networks is proposed. This method performs topology partitioning on a large-scale sensor network and calculates the similarity degree among regional subnetworks. The trend of unknown network security events can be predicted through multitask learning of the occurrence and transmission characteristics of known network security events. Second, in case of lacking regional data, the quantitative trend of unknown regional network security events can be calculated. This study introduces a sensor network security event forecast method named Prediction Network Security Incomplete Unmarked Data (PNSIUD) method to forecast missing attack data in the target region according to the known partial data in similar regions. Experimental results indicate that for an unknown security event forecast the forecast accuracy and effects of the similarity forecast algorithm are better than those of single-task learning method. At the same time, the forecast accuracy of the PNSIUD method is better than that of the traditional support vector machine method.

Bionic Spider Web Flexible Strain Sensor Based on CF-L and Machine Learning

“…The application of machine learning is reflected not only in the analysis of data but also in the prediction of data. After analyzing and training the known data of multiregion wireless sensor network by machine learning, Zhang predicted the missing attack data in the target region and successfully predicted the trend of security events in unknown regions . In this paper, we used the prediction and optimization unknown data capabilities of machine learning to successfully predict the preparation parameters of LIG based on known data sets.…”

A new method for the prediction of network security situations based on recurrent neural network with gated recurrent unit

“…The manually extracted features are simple and low-level, which are not robust for the variations of NSS. The shortages of existing algorithms are summarized as follow (He, Zhang, Wang et al , 2016): The existing manually extracted features are from data level, which are the part of external security situations. However, the inner relationships in the network are between the entities of the network.…”