Application of Bayesian Fourier model for predicting travel times for on-demand public transport in Sofia

What is it about?

This paper describes a methodical approach for Module 5 of INNOAIR On-demand Public Transport Platform for route optimization and, in particular, for arrival times predictions. In this paper only one predictive model is presented – Bayesian Fourier model. Dependent variable in the model is the travel time between existing bus stops of public transport of city of Sofia. Analysis reveals components of time series – trend, cycle and seasonality. For the estimation of the trend-cycle component periodogram analysis is used. For the estimation of the seasonality classical additive seasonal decomposition is used. For estimation of impact of weather conditions classical linear regression is used. General algorithm and its steps are presented in details. Results from numeric experiment are shown. Forecast are made and tested using real data for one week. Errors for all models are less than 1 minute – between 15 and 50 seconds. This is very good result for the public transport of city of Sofia.

Featured Image