Exploiting Predictability of Random Vector Functional Link Networks in Forecasting Quality of Service (QoS) parameters of IoT-Based Web Services Data

Abstract

QoS parameters are volatile in nature and possess high nonlinearity, thus making the IoT-based service and recommendation process challenging. An efficient and accurate forecasting model is lacking in this area and needs to be explored. Though an artificial neural network is a prominent option for capturing such nonlinearities, its efficiency is limited by the structural complexity and iterative learning method. The random vector functional link network (RVFLN) significantly reduces the time complexity by randomly assigning input weights and biases without further modification. Only output layer weights are calculated iteratively by gradient methods or non-iteratively by least square methods. It is an efficient algorithm with low time complexity and can handle complex domain problems without compromising accuracy. Motivated by these characteristics, this article develops an RVFLN-based model for forecasting QoS parameter sequences. Two real-world IoT-enabled web service dataset series are used in developing and evaluating the effectiveness of RVFLN-based forecasts in terms of three performance metrics. Experimental results, comparative studies, and statistical tests are conducted to establish the superiority of the proposed approach over four other similar forecasting techniques. The comparative models included are MLR, ARIMA, MLP, and RBFNN. The experimental results revealed that the proposed RVFLN based QoS parameter forecasting gives amended prediction accuracy for majority of the QoS parameters over other forecasts. The superiority of RVFLN is also established through relative worth tests.