Long-term prediction of multiple types of time-varying network traffic using chunk-based ensemble learning

Data-Driven Model for Sliced 5G Network Dimensioning and Planning, Featured With Forecast and “what-if” Analysis

“…Network traffic can be forecasted using the so-called offline methods or online methods [8]. Offline methods collect information about the entire time series and then make forecasts.…”

“…Nowadays, plenty of papers about the 5G network dimensioning refer to Machine Learning (ML) methods. One of the most commonly used ML techniques is neural networks with Long Short-Term Memory (LSTM) units [8]- [10]. However, in the case of LSTM-based neural networks, many authors decide to make one-step-ahead forecast (i.e., prediction for the next time step), which is impractical.…”

“…classification methods, information theory methods, (hidden) Markov models, Gaussian processes, and Poisson models [9], [14], [23], [24]. In the literature, we can also find articles that use supervised methods for forecasting net-work traffic [8], [25]- [27]. These methods include support vector machines, k-nearest neighbors, decision tree, linear regression, AdaBoost and random forest.…”

Model for Enhancing Cloud Computing Resource Allocation Management Using Data Analytics

“…Based on their specific needs, network operators may set parameters regarding overestimation and underestimation, as well as acceptable blocking thresholds. Future work will explore different parameter configurations of the AOBT metric and their effects on the choice of prediction models for multiple types of traffic [10]. An ensemble model based on artificial intelligence for predicting vehicular traffic noise proved to be more robust than single models in dealing with uncertainties when comparing the results obtained by the single models.…”

“…It is important to note that the linear ensemble techniques in the study have limitations in that a lower performance of one of the single models may result in a less robust single mode performance [11][12]. It improves the generalization ability of cloud computing using at least one of the following strategies: (i) averaging strategy, (ii) voting strategy, and (iii) learning strategy [10]. As part of this paper, an averaging strategy for predicting traffic is proposed, which is necessary for allocating resources.…”