Deep survival analysis of searching for on-street parking in urban areas

Abstract

Searching for parking is a significant contributor to urban road congestion leading to additional costs for the driver emerging from the increased time spent traveling and fuel consumption. The present work attempts to model the duration for searching for parking space monitored with smartphone sensing using the widespread parametric and semi-parametric survival models, as well as random survival forests and deep learning survival models. The available dataset consists of more than 48,000 driving trips conducted in the Region of Attica, Greece, and is enriched with exogenous variables, such as population density and land use in each trips’ destination area. Findings reveal that the time of day in which the trip was performed, as well as trip duration and length, significantly affect parking searching duration. In addition, the land use of the destination area appears to be a significant factor for predicting parking searching duration. Although all survival models share similar results in terms of the significance of the parameters, deep survival neural networks noticeably improve the survival time predictions.