Enhanced Belief Function-Based Decision Blending for Detecting Fault in Wireless Sensor Networks

Introduction

Wireless Sensor Networks (WSNs) collect different types of data from a different environment. WSN has a different topology consisting of any number of sensor nodes deployed as different patterns like random, static, dynamic, dense or sparse. WSNs are implemented in different fields according to the rapid development field. WSN is used in data science fields, where a large amount of data is processed. Machine learning algorithms implemented for categories collected large data from sensor nodes. Ensemble learning algorithms are used to aggregate the predicted data of all sensor nodes. This aggregation concept is known as decision fusion, minimizing the wrong model selection by implementing multiclass classification problems (El Hindi et al., 2018). Sensor nodes in WSN are communicated with others directly or indirectly. The head node merges the collected data and decisions are made accordingly. The head sensor node must be at the center position of the network otherwise at decentralized case data accuracy is a problem. Local sensor nodes cannot deal with data accuracy problems. Therefore, fusion rules are implemented to fuse the reliable data. To analyze the local sensor data detection and the false alarm to confirm the decision fusion rule proposed for the data accuracy (Sriranga et al., 2018). In (Samet, Lefevre, Yahia, 2014) belief function fusion approaches an estimation classifier used to define the forest image on a classification method to categorize problems of images and overcome methods. Here to find distance between targets distance estimation model classifier used. The collected data from the sensor node is sent to the center head node in binary form. For accurate data all the data compare with threshold value set by the system and reach at the final decision implementing optimal decision. Optical fusion rules are used to decode and forward relaying for different channels. There are different techniques based on intuitionist algorithms i.e., data based IFS and weight based TOPSIS intuitionist decision algorithm on fuzzy. Fusion rule used to find the amount of required energy and working condition based on statistical channel information and on-off keying (Zhang et al., 2017). Here the Gaussian mixture is used for multi-input-output decision fusion based on energy detection over different classes of fading distribution methods. Multiple classifiers and techniques are used to evaluate the problems related to energy dissipation. Deep learning algorithms and classifiers are used here. Initially the decision tree, Fourier based, was used as a basic classifier to perform on data loss, aggregation error, calibration fault in WSN (Das et al., 2021; De et al., 2020). There are different types of faults in WSN which can be detected using machine learning techniques. Types of faults defined according to sensing data.

• (a). Offset fault: Calibration’s error and actual data added in expected data at sensing unity called offset fault.

• (b). Gain fault: Sensed data rate varies in specific time known as gain fault.

• (c). Stuck-at fault: Null valued data found at sensing and that not changed with time known as stuck at fault.

• (d). Out of bounds: Data value will be different and that out of boundary from normal value.

According to different types of faults, the machine learning algorithm applied to improve accuracy in the fusion of belief functions. Complex Basic Belief Assignment (BBA) approach used to upload the BBA structure to the fusion center using each sensor node. Here used a different method to get a better result in fault detection i.e., Enhanced Support Vector Machine (ESVM), Enhanced K-Nearest Neighbor (EKNN), Enhanced Extreme Learning Machine (EELM) and Enhanced Recurrent Extreme Learning Machine (ERELM) (Das et al., 2021).