Classification And Regression Tree (CART) based resource allocation scheme for Wireless Sensor Networks

Abstract

The 21st century is the era of smart sensors, intelligent computations, and communication technologies. Wireless Sensor Networks (WSN) play a vital role in performing many remote applications without human intervention. WSNs are an integral part of an ubiquitous computing and Internet of Everything (IoE). A WSN is a decentralized network comprised of several tiny but powerful sensor nodes. Sensor nodes are restricted in terms of battery life, communication range, bandwidth, processing latency, and memory. Effective usage of WSN resources is a challenging task to enhance network lifespan, increase throughput, reduce computational delay and minimize control overheads. Several intelligent strategies are proposed to improve WSN performance and enhance the lifespan of the network by adopting intelligent resource management schemes. In WSN, effective and intelligent resource management involves resource discovery, resource scheduling, and resource allocation. In this paper, a Classification and Regression Tree (CART) supervised machine learning algorithm is used to deal with incomplete information about the network i.e., uncertainty and dynamicity for effective resource allocation. The scheme operates as follows: The k-means clustering algorithm is applied to the network, clusters are formed, the Cluster Head (CH) is selected, and the k-NN algorithm is applied to find the number of neighbor nodes i.e., Cluster Members (CM) in the cluster. The attributes of CH like distance from base station, degree of connectivity, congestion rate, data type and size aggregated at CH after performing a task and channel quality are calculated. Aggregation and classification of CM and CH attributes (data sets) use intelligent search and feature selection algorithms. The data sets are then processed for the training and prediction phases. A decision tree model is built using target attribute, which is resource (bandwidth) allocation. A heat map and confusion matrix are generated and a performance evaluation of the proposed scheme is done. Simulation results show that the performance of the proposed CART based resource allocation approach is better as compared with the “Linear Regression (LR), Iterative Dichotomiser 3 (ID3), and Neural Network (NN)” schemes in terms of resource allocation accuracy, allocation computational delay, data transmission efficiency, etc.