Abstract
One of the core challenges in open-plan workspaces is to ensure a good level of concentration for the workers while performing their tasks. Hence, being able to infer concentration levels of workers will allow building designers, managers, and workers to estimate what effect different open-plan layouts will have and to find an optimal one. In this research, we present an ambient-physical system to investigate the concentration inference problem. Specifically, we deploy a series of pervasive sensors to capture various ambient and physical signals related to perceived concentration at work. The practicality of our system has been tested on two large open-plan workplaces with different designs and layouts. The empirical results highlight promising applications of pervasive sensing in occupational concentration inference, which can be adopted to enhance the capabilities of modern workplaces.
Introduction
Open-plan offices are a double-edged sword: while they promote collaboration and reduce real estate costs, they are frequently criticized for distractions, noise, and a lack of privacy. These factors significantly impact productivity, yet measuring how and when they affect an individual's concentration in real-time has historically been difficult without intrusive monitoring.
This research addresses the challenge of "concentration inference"—automatically determining a worker's focus level based on their environmental surroundings and physical behavior. By moving away from subjective assessments toward data-driven insights, this system provides a foundation for optimizing office layouts and adjusting ambient conditions to help workers maintain their "zone" of focus.
Methodology
The researchers developed a multi-layered system architecture that integrates two primary data streams: ambient environmental data and physical activity data.
System Architecture
The system utilizes a Passive Sensing Pool consisting of Netatmo Weather Stations to collect ambient data (temperature, humidity, CO2, noise, and pressure) and a custom smartphone application, OpenSense, to gather physical data (accelerometer, gyroscope, and pedometer) at a 50 Hz sampling rate.
Data Processing and Fusion
To handle the differing temporal resolutions of these data streams, the system employs a sliding window approach. Ambient features are extracted over 30-minute windows, while physical movement features use 5-minute windows. These disparate streams are aligned by timestamp to create a combined "ambient-physical" feature vector.
For classification, the study tested eight machine learning models. The CatBoost algorithm, a gradient-boosting framework, emerged as the top performer due to its efficiency with categorical data and imbalanced datasets.
Experimental Setup
The system was deployed over four weeks across two distinct engineering consultancy sites in Australia:
- Site-1: A traditional single-floor open-plan office.
- Site-2: A modern, multi-layered space featuring a central void and various functional zones.
Thirty-one corporate workers participated in the study, providing ground truth by self-reporting their concentration levels twice daily (10:00 AM and 3:30 PM) using a 5-point Likert scale via the OpenSense app.
Key Results and Insights
The study's findings underscore the importance of data fusion. Combining ambient and physical data (A+P) led to a substantial accuracy increase, often 10-30% higher than using physical data alone.
- Performance Milestones: In Site-1, the system achieved a peak accuracy of 97.16% for morning sessions. In the more complex Site-2 environment, accuracy reached 76.07%.
- Environmental Drivers: High CO2 density and temperatures outside the 21.5–22.5°C range were strongly correlated with lower concentration levels. Magnetometer and accelerometer readings were also among the top indicators of focus.
- The Impact of Layout: Office "topology" significantly influenced focus. While a high number of formal meetings negatively impacted workers in Site-1, informal interactions in the modern Site-2 layout were associated with better median concentration levels.
Conclusion
This research demonstrates that pervasive sensing can effectively infer occupational concentration in diverse workplace environments. While the current model relies on periodic self-reports for ground truth, future work aims to develop even less obtrusive data collection methods and transition toward sequential time-series modeling to better capture the dynamic nature of workplace focus.
Summary updated