ParkMapSense: A model for parking occupancy mapping and prediction in smart cities

Abstract

Urban mobility challenges have intensified with the rapid expansion of private vehicle fleets. To mitigate parking scarcity in central areas, municipalities often implement Metered Parking Zones to encourage vehicle turnover. These digitized systems generate historical data following the Arrive-Stay-Leave (ASL) pattern, enabling occupancy mapping and predictive analysis. This study introduces ParkMapSense, a computational model designed to generate occupancy maps and forecast parking availability. The model was validated using operational data from Novo Hamburgo, RS (Brazil), covering November 2022 to May 2024, comprising 1.4 million inspections and 1.3 million parking activations across 2,052 parking spaces. Occupancy was aggregated into 10-minute intervals, and prediction algorithms based on Global, 7-day, and 28-day moving averages were applied, segmented by weather conditions and day type. Rainfall data were integrated to analyze behavioral variations. Evaluation metrics showed R 2 scores of 0.86 for weekdays without rain, 0.74 with rain, and 0.81 on Saturdays in any weather, confirming the model's ability to capture temporal patterns. The main contribution of this study lies in the occupancy mapping algorithm, which transforms metered parking data into structured time-series occupancy maps, and in the weather-aware segmentation strategy, which serves as lightweight baselines within an extensible structure designed to support future integration of other prediction algorithms. ParkMapSense demonstrated low computational cost and operational feasibility for medium-sized cities, offering valuable insights for urban mobility planning and enforcement optimization.