Prediction of Self-similar Behavior Internet user’s Traffic Data through ANN and Multilinear Regression Models

Resumen

One of the major challenges in network management is effectively managing and scheduling the arrival patterns of internet users across various web centers. In this study, the Hurst method is employed to analyse the self-similarity of online user traffic, helping to identify performance degradation in page loading during periods of high traffic. A linear regression model is utilized to fit the observed data and to formulate a mathematical equation for forecasting user arrival patterns. Furthermore, an Artificial Neural Network (ANN) with a multilayer perceptron architecture is implemented to predict online user behavior more accurately. The study also evaluates and compares the prediction accuracies of both models. Traditional web traffic prediction models, such as Poisson and Markov-based approaches, fail to adequately capture the self-similar and bursty nature of internet user arrivals. Moreover, most existing studies focus primarily on traffic characterization rather than predictive modeling, with limited comparative analysis between statistical and machine learning approaches.