Deep or statistical: an empirical study of traffic predictions on multiple time scales

Abstract

Traffic prediction aims to forecast the future traffic level based on past observations. In this paper, we conduct an empirical study of traffic prediction for a campus trace on different time scales and get the following conclusions: 1) deep learning performs well on coarser time scales; 2) with a finer-granularity of time or insufficient data, statistical and regressive models outperform; 3) For a one-week trace, the granularity of 5 minutes has the strongest predictability.