Examining parking preferences with private autonomous vehicles using random forest and logit models: the case of commuters to central Melbourne

Private autonomous vehicles (PAVs) have the potential to significantly reduce the demand for parking spaces and costs in urban centres through their self-driving capabilities to find free or cheaper parking options elsewhere. Yet, limited research has examined how factors such as residential location, residential parking type, and workplace parking payment type (e.g., self-paid or employer-paid) influence PAV parking choices and associated empty-cruising vehicle-kilometres travelled (VKT). This study addresses these gaps using survey data collected from 526 commuters driving to Central Melbourne, Australia. We employ both random forest (RF) and multinomial logistic regression (MNL) models to investigate factors influencing different parking options. The MNL results show that commute time, household composition, residential parking type, region, age, housing type, and walking time are key determinants of preferences for sending PAVs home. Meanwhile, education, household composition, region, and car ownership influence preferences for free suburban parking. These results align with RF model importance rankings, where commute time (19 %) and household composition (11 %) were the top predictors. Additionally, sending vehicles home could generate 13 % more VKT compared to current commuter patterns. These findings have implications for managing future CBD parking supply, regulating empty cruising, and shaping land use and pricing strategies for PAV-era mobility.