A Comparison of ARIMAX, VAR and LSTM on Multivariate Short-Term Traffic Volume Forecasting

Traffic volume forecasting is a key objective in Intelligent Transportation Systems (ITS) since its importance for both the general public and authorities in decision making, optimizing navigation strategies and avoid traffic congestions. Various research projects have been conducted for identifying...

Full description

Saved in:
Bibliographic Details
Published in:Proceedings of the XXth Conference of Open Innovations Association FRUCT 2021-01, Vol.28 (2), p.564-570
Main Authors: Bhanuka Dissanayake, Osanda Hemachandra, Nuwan Lakshitha, Dilantha Haputhanthri, Adeesha Wijayasiri
Format: Article
Language:English
Subjects:
arimax
lstm
multivariate time series forecasting
short term traffic forecasting
var
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Traffic volume forecasting is a key objective in Intelligent Transportation Systems (ITS) since its importance for both the general public and authorities in decision making, optimizing navigation strategies and avoid traffic congestions. Various research projects have been conducted for identifying the best approach to solve that issue. This paper proposes a comparison of statistical learning models, Vector Auto Regression, ARIMAX and a deep learning model, LSTM neural network, in the context of multivariate short-term (24 hours) time series forecasting using traffic volume, speed, and average waiting time, integrating weather attributes in Austin city, Texas. Models were evaluated using rolling forecast origin method for three main feature sets generated utilizing feature selection. VAR model produced the best performance with an accuracy of 91.459% and proved to be used successfully in short term traffic forecasting in ITS applications.
ISSN:2305-7254
2343-0737
DOI:10.5281/zenodo.4514955