An improved Prophet emergency traffic-flow prediction model

Abstract

Emergencies are rare and random, but cause dramatic changes in traffic volumes whenever they occur. It is therefore very challenging to predict emergency traffic volumes accurately. The aim of this paper is to provide a practical methodology for predicting traffic volumes during emergencies; this is done using the Prophet model and by improving the event function of the model. The proposed approach isolates the event impacts through time series decomposition techniques; it allows the model to add points in time when traffic flow changes abruptly and to incorporate external factors useful to adapt to the specific background imposed by the emergency. The main data used in the paper were from the daily traffic volume data set published on the Luxembourg Open Data platform. These data were collected over a period from January 2017 to December 2021. The data set covers the period of impact of two emergencies. The proposed method and four comparator models were applied to the second emergency period. The results show that the proposed method can accurately predict unconventional changes caused by emergencies and, by way of a comparative analysis, that it has better prediction accuracy with real data than the other comparator models under the same attribute conditions.