Autoregressive integrated moving average

Mathematics

Parametric statistics

Statistics

Machine learning

Ensemble forecasting

Time series

Econometrics

Artificial intelligence

Parametric model

Bootstrapping (finance)

Computer science

Ensemble learning

Transportmetrica A Transport Science

There are numerous studies on traffic volume prediction, using either non-parametric or parametric methods. The main shortcoming of parametric methods is low prediction accuracy. Non-parametric methods show higher prediction accuracy, but they are criticised due to lack of support from theory. The innovation of this paper is to combine bootstrap with the conventional parametric ARIMA model with the aim of improving prediction accuracy while maintaining theory adherence. The outcome of this process is an ensemble of ARIMA models (E-ARIMA) where each model is developed using a random subsample of data. The validity of the proposed model is examined by comparing E-ARIMA with ARIMA and Long Short-Term Memory (LSTM) as representatives for parametric and non-parametric methods respectively. One year of traffic count data on four main arterial roads in Sydney, Australia is used for calibration and validation purposes. The results suggest that creating an ensemble of models improves prediction accuracy.

Ensemble of ARIMA: combining parametric and bootstrapping technique for traffic flow prediction