Traffic Burst Prediction in Radio Access Network with Machine Learning
Jing Jin · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2016
Motivated by the expansion of mobile data traffic, there is an increasingdemand for better allocation of radio resources in the radio access network(RAN). Recently, interest has shifted towards predictive resource allocationtechniques, which would enable a more intelligent RAN. A promising solutionfor developing predictive resource allocation techniques is to combine radioresource allocation algorithms with prediction algorithms based on Machinelearning (ML). In this project, the prediction of data traffic in RAN withML techniques is studied, with the objective to incorporate the predictor incarrier aggregation. The traffic predicted in this project is at the burst levelwhich is an aggregation of several consecutive packets, and the focus is onsupervised classification algorithms. The volume of the burst, burst durationtime and the time gap between two bursts are predicted. The performanceof prediction is evaluated with the receiver operating characteristic curve.