Feature-Selection Based Data Prioritization in Mobile Traffic Prediction Using Machine Learning
Yoshinobu Yamada, Ryoichi Shinkuma, Takehiro Sato, Eiji Oki · 2018
Recently, the demand for realtime and accurate prediction of mobile traffic has been growing in traffic engineering and dynamic resource allocation that work to handle increased mobile data traffic. However, most conventional prediction techniques assumed that traffic logs at every unit time at every base station are perfectly available. This assumption is critical in realtime mobile traffic prediction because the volume of traffic log data collected at base stations is huge and they compete bandwidth with normal user application traffic when they are sent from base stations to the server that performs prediction. Therefore, in realtime mobile traffic prediction, we should consider the condition in which the bandwidth ensured for forwarding traffic log data is limited. In this paper, we propose a method that prioritizes traffic log data in the basis of the contribution to prediction accuracy; each base station sends more important traffic log data to the server with higher priority. The importance of each data entry of traffic log data means how much prediction accuracy would degrade if the entry is missing. The proposed method enables us to reduce the volume of traffic log data sent from base stations to the server while maintaining prediction accuracy at the sufficient level. Our simulation study using a real dataset of mobile-traffic measurement validates our method in terms of prediction accuracy under the limitation of available traffic log data.