A network traffic flow prediction with deep learning approach for large-scale metropolitan area network

Wang, Weitao; Bai, Yuebin; Yu, Chao; Gu, Yuhao; Feng, Peng; Wang, Xiaojing; Wang, Rui

doi:10.1109/noms.2018.8406252

Cited by 39 publications

(23 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The algorithms correctness and completeness were proven, and their complexities were analyzed. An interesting juxtaposition can be done with our model and a proposed approach applied to a real-case big network of Wang et al (2018), that applies a deep learning perspective to a connected traffic flow prediction. While the traffic flow prediction smartly intertwines efficient learning algorithms, the un-distributed network still has a high latency and overhead in comparison with our distributed network model, that needs less data to discover important features and breaches of them in the network.…”

Section: Discussionmentioning

confidence: 99%

Managing large distributed dynamic graphs for smart city network applications

2019

View full text Add to dashboard Cite

Smart cities and traffic applications can be modelled by dynamic graphs for which vertices or edges can be added, removed or change their properties. In the smart city or traffic monitoring problem, we wish to detect if a city dynamic graph maintains a certain local or global property. Monitoring city large dynamic graphs, is even more complicated. To treat the monitoring problem efficiently we divide a large city graph into sub-graphs. In the distributed monitoring problem we would like to define some local conditions for which the global city graph G maintains a certain property. Furthermore, we would like to detect if a local city change in a sub-graph affect a global graph property. Here we show that turning the graph into a non-trivial one by handling directed graphs, weighted graphs, graphs with nodes that contain different attributes or combinations of these aspects, can be integrated in known urban environment applications. These implementations are demonstrated here in two types of network applications: traffic network application and on-line social network smart city applications. We exemplify these two problems, show their experimental results and characterize efficient monitoring algorithms that can handle them.

show abstract

Section: Discussionmentioning

confidence: 99%

Managing large distributed dynamic graphs for smart city network applications

2019

View full text Add to dashboard Cite

show abstract

“…Previous studies [7]- [9] investigated machine learningbased traffic prediction for realizing proactive control. The authors in [7] utilized artificial neural networks, while those in [8] utilized deep learning, which is a type of neural network. However, a neural network-based computational process generally requires a large amount of training data and its learning procedure is time consuming owing to its complex learning model.…”

Section: Resource Adjustmentsmentioning

confidence: 99%

“…The authors in [7]- [14] investigate machine-learning based traffic prediction for realizing proactive control. The authors in [7], [10], [12] utilize artificial neural networks, while those in [8], [11] utilize deep learning. However, its model is so complicated, and thus, makes it difficult for humans to perform posterior analysis after learning processes.…”

Section: Network Traffic Prediction Using Machine Learningmentioning

confidence: 99%

Sparse Regression Model-Based Relearning Architecture for Shortening Learning Time in Traffic Prediction

Hirayama

Miyazawa

Jibiki

et al. 2021

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

Network function virtualization (NFV) enables network operators to flexibly provide diverse virtualized functions for services such as Internet of things (IoT) and mobile applications. To meet multiple quality of service (QoS) requirements against time-varying network environments, infrastructure providers must dynamically adjust the amount of computational resources, such as CPU, assigned to virtual network functions (VNFs). To provide agile resource control and adaptiveness, predicting the virtual server load via machine learning technologies is an effective approach to the proactive control of network systems. In this paper, we propose an adjustment mechanism for regressors based on forgetting and dynamic ensemble executed in a shorter time than that of our previous work. The framework includes a reducing training data method based on sparse model regression. By making a short list of training data derived from the sparse regression model, the relearning time can be reduced to about 57% without degrading provisioning accuracy.

show abstract

“…Similarly, the study in [6] proposes a cognitive management framework with unsupervised deep learning and probabilistic generative models for network optimization. The paper in [7] proposes a prediction model with deep learning for internet traffic flow forecast in real-time. To build Self Organizing Networks (SONs), the paper in [8] leverages cellular mobile data to cluster the Base Stations (BSs) using unsupervised learning approaches so that inter-cluster handover rates can be reduced.…”

Section: A Related Workmentioning

confidence: 99%

A Machine Learning Based Management System for Network Services

Türk¹,

Zeydan²,

Bilgin³

2019

2019 International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)

View full text Add to dashboard Cite

Providing high quality and uninterrupted network service is becoming crucial for service providers. In this paper, we present an approach to quantify and indicate service quality based on the topology state transitions from the perspective of network service provider. Building our model as a Finite State Machine (FSM), we show novel application of machine learning (ML) classification algorithms to classify appropriate states for undefined input alphabets in FSM. In other words, we implement ML algorithms to extract both service states and possible root causes of service degradation only from measured certain Key Performance Indicator (KPI) values that are observed directly through network elements. We have implemented our network topology state classification approach using the dataset obtained in Graphical Network Simulator 3 (GNS3) simulation environment, and performed measurements to evaluate classification accuracy of different algorithms. Additionally, we have identified priority of relevant KPIs impacting the service quality. Our results indicate that network topology state changes can be classified up to 88% accuracy and F1 scores using ensemble learning methods such as Gradient Boosting Classifiers.

show abstract

A network traffic flow prediction with deep learning approach for large-scale metropolitan area network

Cited by 39 publications

References 29 publications

Managing large distributed dynamic graphs for smart city network applications

Managing large distributed dynamic graphs for smart city network applications

Sparse Regression Model-Based Relearning Architecture for Shortening Learning Time in Traffic Prediction

A Machine Learning Based Management System for Network Services

Contact Info

Product

Resources

About