Congestion Detection in Wireless Sensor Networks Based on Artificial Neural Network and Support Vector Machine

A wireless sensor network (WSN) monitors a certain phenomenon through interconnecting voluminous quantity of interconnected sensor nodes that are implemented in the sensing space and event. Transmission reliability is the key contributing factor in of WSN. Network latency, data packet loss, throughput reduction, and low energy efficiency is caused by a congested network. To solve the network congestion issue, a proposed strategy of congestion detection using ANN (artificial neural network) in comparison to SVM (support vector machine) is presented in this paper. The number of sensor nodes, traffic rate, and node retention are the parameter used. These were generated from the WSN setup in the lettuce smart farm located in Rizal, Philippines. Network Simulator 2 (NS-2) is used for network design, traffic generation, and data gathering. The 3-layer feedforward ANN is optimized using scaled conjugate gradient with sigmoidal function as output activation. SVM is configured using box constraint of 564.22 and kernel scale of 0.1475. There is 0.00776625 CE performance (cross-entropy) and 98.8% accuracy through network training. Conversely, SVM shows 95.18% accuracy. Thus, the developed ANN-based WSN congestion detection model with +3.62% accuracy is an effective tool for agricultural wireless network where array of sensors for crop and environmental monitoring is connected.

I. Introduction

Internet of things (IoT) and Wireless sensor nodes (WSNs) are merged in a massiv earray of nodes containing sensor readers that are listlessly implemented and contending for a sensing event either inside a sensing region or approximately close to it [1]. Monitoring events and recording data in a region limited to human potentiality is one of the advantages of WSNs. Each nodes are usually placed in a region where people would not be comfortably to stay at for continuous monitoring. Individual sensor node has the feature for collecting data and transmitting it back to the router. Data are received via a regional node to node architecture through the sink. On the other hand, the Internet or a global satellite may be used by the sink to transport data into and from the task manager node [2]. The position of sensor nodes need not be contemplated because sensor network architecture must me embedded with self-organizing map algorithm. By default, a sensor network is self- reliant but interdependent on nature. Provided that sensor nodes has internal and system-wide autonomous processing capabilities, these sensor nodes transmits only partially computed and required data to the end user resulting to lesser congestion. This is an integration approach to be considered in constructing a WSN, configuring sensor nodes to process and transmit only the essential data.