Comparing Aggregated and Separate Models for Crowd Density Forecasting

• Type: Article in International Conference Proceedings Book
• Date: 2025

Authors

Sardinha, I (Author) Other; Ana Paula Rocha (Author) FEUP; Rosaldo J. F. Rossetti (Author) FEUP

Conference proceedings International

Title: PROGRESS IN ARTIFICIAL INTELLIGENCE, EPIA 2024, PT I

Pages: 386-397

23rd EPIA Conference on Artificial Intelligence-EPIA

Viana do Castelo, PORTUGAL, SEP 03-06, 2024

Indexing

ISI Web of Knowledge - 0 Citations

Scopus - 0 Citations

Other information

Authenticus ID: P-017-DC1

DOI: 10.1007/978-3-031-73497-7_31

Abstract (EN): Crowd density forecasting applied to smart cities is a task that can provide essential data for a city planner to use in different applications either for long-term optimization strategies, such as building infrastructure, or for daily operations, such as traffic network mapping, security allocation, and organizing events. Predicting crowd density can be seen as a time-series problem that analyses spatio-temporal data on a network representing the number of people in a specific space and the flow between these spaces over time, which is the factor that changes the density. Although modern machine learning algorithms efficiently predict spatial and temporal data, joining the two can still be challenging. This work compares approaches for predicting temporal data and adding spatial information to predict spatio-temporal data. This is done by implementing forecasting methods individually for zones of a city and then single models for the whole city and empirically comparing their results. Different machine learning and deep learning methods are created using mobile communications data from a city in Portugal. The best results were obtained by Graph Neural Networks (GNN), an architecture of deep learning optimized for graph-structured data, such as geographical inputs. It obtained a mean absolute error of 0.5 compared to 31.9 from the second-best model. Individual zones from other methods had comparable results; however, only GNNs could keep consistent results throughout the city.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 12