ParkIQ A Predictive Framework for Urban Parking and Accountable Kerbside Management

Abstract

Parking information is often split between occupancy sensors, payment services and local restriction records. This makes it difficult to tell drivers where they may park, estimate whether a space will remain available, and explain an enforcement decision from a consistent record. This paper proposes ParkIQ, an engineering framework that connects these tasks through a shared representation of individual parking bays. The design combines six layers: sensing, a geospatial bay model, time-dependent restrictions, data fusion, occupancy forecasting and user services. A temporal convolutional network and a gradient-boosted model are proposed for probabilistic forecasts at 15-minute intervals up to 90 minutes ahead. An associated enforcement workflow preserves evidence for authorised review, while a separate commercial model supports bookings and service provision. The contribution is a system specification and prospective evaluation protocol; no trained-model results, operational savings or deployment outcomes are reported. The protocol compares the proposed ensemble with simpler baselines, separates training from future observations, and tests calibration, sensor failures and transfer between locations. It also distinguishes evidence integrity from the correctness and legal validity of an enforcement decision. The framework provides a testable basis for investigating whether shared bay information can improve parking guidance and operational accountability. Its value will depend on data quality, lawful access to information, user participation and evidence from a controlled pilot.