Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Place with Intention: An Empirical Attendance Predictive Study of Expo 2025 Osaka, Kansai, Japan

Xiaojie Yang, Dizhi Huang, Hangli Ge, Masahiro Sano, Takeaki Ohdake, Kazuma Hatano, Noboru Koshizuka

TL;DR

This work addresses daily attendance forecasting for a six-month international exposition by leveraging reservation dynamics as a direct proxy for attendance intentions. It introduces a lightweight Transformer-based framework with an encoder–decoder or decoder-only configuration and an adaptive fusion module that decomposes predictions into a baseline, a gated reservation signal, and a smoothed kernel. Empirical results show reservation dynamics yield strong near-term predictive power (high correlations with attendance) and that modeling the East and West gates separately improves accuracy, with ablations confirming the value of the encoder–decoder architecture, inverse-style temporal embedding, and adaptive fusion. The approach provides a practical, data-efficient tool for crowd management, resource planning, and safety at large-scale events, while acknowledging reliance on reservation system stability and the need for broader generalization tests.

Abstract

Accurate forecasting of daily attendance is vital for managing transportation, crowd flows, and services at large-scale international events such as Expo 2025 Osaka, Kansai, Japan. However, existing approaches often rely on multi-source external data (such as weather, traffic, and social media) to improve accuracy, which can lead to unreliable results when historical data are insufficient. To address these challenges, we propose a Transformer-based framework that leverages reservation dynamics, i.e., ticket bookings and subsequent updates within a time window, as a proxy for visitors' attendance intentions, under the assumption that such intentions are eventually reflected in reservation patterns. This design avoids the complexity of multi-source integration while still capturing external influences like weather and promotions implicitly embedded in reservation dynamics. We construct a dataset combining entrance records and reservation dynamics and evaluate the model under both single-channel (total attendance) and two-channel (separated by East and West gates) settings. Results show that separately modeling East and West gates consistently improves accuracy, particularly for short- and medium-term horizons. Ablation studies further confirm the importance of the encoder-decoder structure, inverse-style embedding, and adaptive fusion module. Overall, our findings indicate that reservation dynamics offer a practical and informative foundation for attendance forecasting in large-scale international events.

Place with Intention: An Empirical Attendance Predictive Study of Expo 2025 Osaka, Kansai, Japan

TL;DR

This work addresses daily attendance forecasting for a six-month international exposition by leveraging reservation dynamics as a direct proxy for attendance intentions. It introduces a lightweight Transformer-based framework with an encoder–decoder or decoder-only configuration and an adaptive fusion module that decomposes predictions into a baseline, a gated reservation signal, and a smoothed kernel. Empirical results show reservation dynamics yield strong near-term predictive power (high correlations with attendance) and that modeling the East and West gates separately improves accuracy, with ablations confirming the value of the encoder–decoder architecture, inverse-style temporal embedding, and adaptive fusion. The approach provides a practical, data-efficient tool for crowd management, resource planning, and safety at large-scale events, while acknowledging reliance on reservation system stability and the need for broader generalization tests.

Abstract

Accurate forecasting of daily attendance is vital for managing transportation, crowd flows, and services at large-scale international events such as Expo 2025 Osaka, Kansai, Japan. However, existing approaches often rely on multi-source external data (such as weather, traffic, and social media) to improve accuracy, which can lead to unreliable results when historical data are insufficient. To address these challenges, we propose a Transformer-based framework that leverages reservation dynamics, i.e., ticket bookings and subsequent updates within a time window, as a proxy for visitors' attendance intentions, under the assumption that such intentions are eventually reflected in reservation patterns. This design avoids the complexity of multi-source integration while still capturing external influences like weather and promotions implicitly embedded in reservation dynamics. We construct a dataset combining entrance records and reservation dynamics and evaluate the model under both single-channel (total attendance) and two-channel (separated by East and West gates) settings. Results show that separately modeling East and West gates consistently improves accuracy, particularly for short- and medium-term horizons. Ablation studies further confirm the importance of the encoder-decoder structure, inverse-style embedding, and adaptive fusion module. Overall, our findings indicate that reservation dynamics offer a practical and informative foundation for attendance forecasting in large-scale international events.
Paper Structure (17 sections, 12 equations, 4 figures, 4 tables)

This paper contains 17 sections, 12 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Dataset and Analysis
  4. Entrance Data
  5. Reservation Data
  6. Analysis and Summary
  7. Methodology
  8. Problem Setup
  9. Temporal Embedding
  10. Model Architecture
  11. Adaptive Fusion
  12. Objective Function
  13. Experiments
  14. Experiment Settings
  15. Overall Performance
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Heatmap of correlations between daily visits and lagged visits (1–10 days) across months (May–August). The figure shows that, besides the strong correlation with the previous day (lag 1), the 7-day lag generally exhibits the second-highest correlation, indicating a clear weekly periodicity in visit patterns
  • Figure 2: Heatmap of correlations between daily visit counts and reservation counts from up to 10 days earlier (lag 0–10), computed across months (May–August). The correlations are generally very high and increase as the reservation date approaches the actual visit day, highlighting the strong predictive relationship between reservations and real attendance
  • Figure 3: Case studies of reservation dynamics and daily changes for different dates. Subplots illustrate: (a) a regular weekday (June 5), (b) a weekend (June 7), (c) a rainy day (June 10), (d) the Blue Impulse air show day (July 12), and (e) a fireworks festival day (August 23). The left panels compare cumulative reservations with actual attendance, while the right panels depict daily reservation changes, highlighting how weather and special events strongly influence booking behavior.
  • Figure 4: Model architecture overview. The encoder processes historical observations to extract contextual features $o_{\text{enc}}$, while the decoder integrates future-aware inputs and temporal markers to generate $o_{\text{dec}}$. The decoder output is then passed to the Adaptive Fusion module for final prediction $\hat{y}$.