Abstract
The technology described here is about predicting future network traffic conditions in a network of computing devices, using a machine learning model, such as a long short-term memory (LSTM) model. An LSTM model is trained on network traffic data, which characterizes or quantifies traffic conditions on different links or devices of a network. The output of the trained model is a prediction of traffic conditions over the network within a future time window, for example within the next minutes, hours, days, or longer. The model is trained with a weighted version of the root mean squared error (RMSE) error function, to compensate for potential underestimation of network conditions. This compensation improves the likelihood that a traffic controller does not reroute data or provision computing resources insufficiently in response to the model prediction. The model can receive traffic data continuously, instead of in temporal snapshots, allowing the model to respond more quickly to changing traffic patterns, as well as provide additional training to further improve the model.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
N/A, "Network Egress Load Prediction using Artificial Neural Network (ANN) Models", Technical Disclosure Commons, (March 29, 2024)
https://www.tdcommons.org/dpubs_series/6833