Predicting Parking Availability in Singapore with Cross-Domain Data: A New Dataset and A Data-Driven Approach

TL;DR

This work tackles real-time, cross-domain prediction of Parking Availability (PA) across Singapore. It introduces SINPA, a large-scale dataset with 1,687 parking lots and cross-domain features, and proposes DeepPA, a scalable spatio-temporal framework that combines a Graph Cosine Operator for efficient spatial modeling with a Temporal Learning Block that uses causal Multi-Head Attention. DeepPA achieves a 9.2% MAE improvement for up to 3-hour forecasts over strong baselines and supports real-time deployment via a web platform, making it practical for drivers and urban planners. The public release of SINPA and the accompanying code underscores the study's significance for smart city parking management and cross-domain forecasting research.

Abstract

The increasing number of vehicles highlights the need for efficient parking space management. Predicting real-time Parking Availability (PA) can help mitigate traffic congestion and the corresponding social problems, which is a pressing issue in densely populated cities like Singapore. In this study, we aim to collectively predict future PA across Singapore with complex factors from various domains. The contributions in this paper are listed as follows: (1) A New Dataset: We introduce the \texttt{SINPA} dataset, containing a year's worth of PA data from 1,687 parking lots in Singapore, enriched with various spatial and temporal factors. (2) A Data-Driven Approach: We present DeepPA, a novel deep-learning framework, to collectively and efficiently predict future PA across thousands of parking lots. (3) Extensive Experiments and Deployment: DeepPA demonstrates a 9.2% reduction in prediction error for up to 3-hour forecasts compared to existing advanced models. Furthermore, we implement DeepPA in a practical web-based platform to provide real-time PA predictions to aid drivers and inform urban planning for the governors in Singapore. We release the dataset and source code at https://github.com/yoshall/SINPA.

Figures (6)

• Figure 1: Distribution of 1,687 parking lots throughout Singapore. Each lot reports real-time parking availability every 15 minutes.
• Figure 2: The framework of our proposed DeepPA. Initially, historical PA readings and spatio-temporal information are encoded and mapped into hidden states, respectively. Subsequently, PA will interact with spatio-temporal information in two sequential stages: (a) Spatial Learning Block (SLBlock) captures spatial dependence among parking lots, in which we introduce a novel structure Graph Cosine Operator (GCO) to enhance interaction efficiency; (b) TLBlock simultaneously investigates the temporal dynamics of PA and temporal information adhering to temporal patterns through Causal Multi-Head Attention (Causal MSA). Finally, DeepPA will integrate the feature space to forecast future PA. Concat: Concatenate. Info:Information
• Figure 3: Effects of SLBlock on MAE.
• Figure 4: Effects of TLBlock on MAE.
• Figure 5: Effects of (a) hidden dimension $C$ and (b) the number of blocks $L$.