The Complexity of Indoor Air Quality Forecasting and the Simplicity of Interacting with It – A Case Study of 1007 Office Meetings

Title of the Paper

The Complexity of Indoor Air Quality Forecasting and the Simplicity of Interacting with It – A Case Study of 1007 Office Meetings

Paper Information

• Field of Study: Human-Computer Interaction (HCI) and Indoor Air Quality (IAQ) Management
• Keywords: Human-Building Interaction, Human-Computer Interaction, Indoor Air Quality, Predictive Model Interaction, Office Meetings

Research Background and Issues

• Research Background:

  • The quality of indoor environments directly impacts health, cognitive performance, and productivity. In developed countries, people spend approximately 90% of their time indoors.
  • Modern airtight buildings often lead to elevated CO₂ levels, which, when exceeding thresholds, can cause fatigue, difficulty concentrating, and even cardiovascular issues.
  • Gases like carbon dioxide (CO₂), due to their relationship with ventilation efficiency, are commonly used as key indicators of indoor air quality.
  • Empirical studies have highlighted the global severity of indoor air quality issues, such as the annual productivity loss of £13 billion attributed to poor office environments in the UK.

• Research Problems:

  1. How can the dynamic changes in CO₂ concentration during office meetings be predicted?
  2. How can interventions be designed to effectively mitigate air quality deterioration without disrupting ongoing activities?
  3. How do users perceive intervention suggestions based on predictive results, and what is their level of acceptance?

• Research Significance:

  1. Improving indoor air quality can enhance short-term health and productivity while reducing long-term health risks.
  2. The COVID-19 pandemic in 2020 underscored the importance of improving air circulation to reduce virus transmission risks.
  3. Predictive methods combined with user participation in controlling indoor environments could foster long-term behavioral and habit changes.

• Related Work:

  • Research in the field of Indoor Air Quality (IAQ) has primarily focused on pollution sources, health impacts, and indoor air management, such as air quality thresholds defined by ASHRAE standards.
  • Some studies in the field of Human-Computer Interaction (HCI) have explored intuitive methods to enhance awareness of indoor environments, though most focus on residential spaces and energy optimization.

Solution

• Research Methods and Proposed Solution:

  • The authors proposed a data-driven predictive method using machine learning algorithms to forecast CO₂ concentration changes during meetings, enabling proactive user intervention.
  • Development of Predictive Model:
    • Dynamic CO₂ concentration data from 1007 meetings were collected, identifying seven distinct CO₂ concentration patterns.
    • Hierarchical clustering methods were used to identify and classify different CO₂ variation patterns.
    • A real-time model was developed based on historical CO₂ concentration data, achieving an accuracy of 87.5% (20 minutes) and 97.6% (5 minutes).
  • Interaction Design:
    • Quantitative risk diagrams (Interruption Risk Diagrams) and user surveys were used to test user responses to various scenarios (risk levels, weather conditions, etc.).
    • Display methods were developed for public devices (e.g., notifications on meeting room screens to indicate optimal times to open windows) or personalized reminders via smart devices.

• Innovations:

  • Proposed a proactive intervention method rather than a passive response to indoor air pollution.
  • Identified typical CO₂ variation patterns for the first time and designed interventions considering user behavior and social dynamics.
  • Combined HCI and predictive modeling to minimize user interruption costs and achieve dynamic optimization in different contexts.

Research Findings

Key Findings

1. Data Analysis:

   • Over 66% of meeting time had CO₂ levels exceeding 600 ppm.
   • Users' subjective perception of CO₂ levels showed weak correlation with actual measurements, highlighting the need for augmented machine sensing capabilities.
   • Seven types of CO₂ concentration variation patterns were identified through data clustering, each with specific characteristics and associated time-based risks.

2. Interaction Solutions:

   • Before Meetings: If a rapid increase in CO₂ is predicted, users are advised to ventilate the room in advance.
   • During Meetings: Based on dynamic CO₂ predictions, alerts are issued via public displays only when risks are high and time is critical.

3. User Survey Results:

   • Users were more willing to accept window-opening suggestions when notified in "high-risk, near-term" scenarios.
   • Weather conditions influenced users' willingness to open windows, while the presence of external participants had minimal impact.
   • Public device displays were found to be the least disruptive method of delivering alerts.

Strengths and Limitations

• Strengths:

  • Compared to traditional air quality monitoring, the predictive method improved responsiveness to CO₂ peaks.
  • The design solution balanced practical needs (health, energy efficiency, and minimal disruption) with user experience.

• Limitations:

  • The study focused solely on naturally ventilated office spaces, limiting the generalizability of results to mechanically ventilated or highly polluted environments.
  • Some variables, such as the actual number of meeting participants, were challenging to collect, potentially limiting the predictive capability.

Limitations and Future Work

1. Extend the predictive model's applicability to mechanically ventilated spaces.
2. Further explore personalized prediction and notification methods to accommodate diverse user behavior preferences.
3. Integrate the solution into existing commercial products like Hilo to evaluate its long-term impact and practical effectiveness.