Improved Sparse PLSR and Run-To-Run Algorithms for Traffic Matrix Prediction in Software-Defined Networks

Abstract

Predicting the correct traffic matrix is crucial in addressing many network issues like routing, availability of networks, clear communication, etc.… In conventional networks, link load measurements are used for better traffic matrix prediction. The precision of a system, however, is small since this method does not evaluate the fundamental series of linear equations governing the problem of traffic prediction. Software-Defined Networks(SDN) offer statistics of certain forms of movements, thus providing different ways of solving issues with the traffic matrix. A network’s performance and complexity can be measured using SDN. In this research work, a deep traffic matrix prediction model is proposed to evaluate the traffic in a better way. Further, the network traffic can be forecasted by using the traffic knowledge of a particular network. Gradient descent is considered to be an iterative machine learning modeling method that helps in reducing errors, optimizing weights, and thereby lowering the cost function. Evaluating the gradient of the negative likelihood equation in calculating the traffic matrix is a challenging part since all state models are composed of the gradient. In the traffic matrix, some methods deliver correct predictions, and some do not. In the proposed work, we have introduced two algorithms, the sparse PLSR and Run-2-Run algorithms for traffic prediction. A comparative study between the proposed algorithms and the already existing algorithms was done based on the training efficiency and performance in traffic prediction.