Unlike cities in the Global North, emerging cities, especially those in Sub-Saharan Africa (SSA), have heavily relied on poorly maintained public transport systems, dominated by paratransit services. From multiple transport surveys, tailored to reflect the existing travel patterns of commuters in Greater Freetown, Republic of Sierra Leone, this study leveraged AI-based and statistical models to calibrate a discrete choice model for conventional paratransit services. It explored the attributes influencing mode choice and quantified the extent to which AI-based models are sensitive to changes in transport policies. The revealed preference (RP) and stated preference (SP) travel-diary data constitute various sociodemographic and trip-related attributes. Two multinomial logit (MNL) models, five deep learning, and seven ensemble machine learning models were calibrated. The findings revealed a strong preference for minibuses and three-wheelers, followed by paratransit buses. The MNL models achieved average prediction accuracies ranging from 60.4% to 80.9%, whereas the AI-based models attained substantially higher accuracies ranging from 90.0% to 94.4%, while preserving comparable behavioural outcomes. The latter models were found to significantly improve the prediction accuracy of the calibrated mode choice model, suggesting the effectiveness of AI-based models in travel behaviour modelling. These models remain useful for accurately forecasting demand across different modes. In terms of attribute importance, travel cost was revealed as the most significant trip-related attribute, followed by bus stop waiting time and region of residence. The models are sensitive to travel cost and bus stop waiting time, highlighting their crucial role in commuter mode choice. The study establishes behavioural relationships, offering the flexibility and scalability needed to support data-driven, user-focused, future-ready, and modernized transport planning in the city.
Turay et al. (Wed,) studied this question.