Interpretable data-driven demand modelling for on-demand transit services

In recent years, with the advancements in information and communication technology, different emerging on-demand shared mobility services have been introduced as innovative solutions low-density areas, including transit (ODT), (MOD) transit, crowdsourced services. However, due to their infancy, there is a strong need understand model demand for these this study, we developed trip production distribution models ODT at Dissemination areas (DA) level using four machine learning algorithms: Random Forest (RF), Bagging, Artificial Neural Network (ANN) Deep (DNN). The data used modelling process were acquired from Belleville’s operational 2016 census data. Bayesian optimalization approach was find optimal architecture of adopted algorithms. Moreover, post-hoc employed interpret predictions examine importance explanatory variables. results showed that land-use type most important variable model. On other hand, demographic characteristics destination variables revealed higher levels are expected between dissemination commercial/industrial high-density residential land-use. Our findings suggest performance can be further enhanced by (a) locating idle vehicles neighbourhoods (b) spatio-temporal obtained work continuously update operating fleet size.