Three phase classification of an uninterrupted traffic flow: a $k$-means clustering study

Abstract

We investigate the speed time series of the vehicles recorded by a camera at a section of a highway in the city of Isfahan, Iran. Using $k$-means clustering algorithm, we find that the natural number of clustering for this set of data is $3$. This is in agreement with the three-phase theory of uninterrupted traffic flows. According to this theory, the three traffic phases are categorized as {\em free flow} (F), {\em synchronized} (S) and {\em wide moving jam} (J). We obtain the transition speeds and densities at F-S and also S-J transitions. We also apply the Shannon entropy analysis on the speed time series over finite windows, which equips us to monitor in areal time the instant state of a traffic flow.