Repository logo

Future workspace needs flexibility and diversity: A machine learning-driven behavioural analysis of co-working space.

Published version

Repository DOI

Change log


Debnath, Ramit 
Lonsdale, Nathan 


The future of workspace is significantly shaped by the advancements in technologies, changes in work patterns and workers' desire for an improved well-being. Co-working space is an alternative workspace solution, for cost-effectiveness, the opportunity for diverse and flexible design and multi-use. This study examined the human-centric design choices using spatial and temporal variation of occupancy levels and user behaviour in a flexible co-working space in London. Through a machine-learning-driven analysis, we investigated the time-dependent patterns, decompose space usage, calculate seat utilisation and identify spatial hotspots. The analysis incorporated a large dataset of sensor-detected occupancy data spanning 477 days, comprising more than 140 million (145×106) data points. Additionally, on-site observations of activities were recorded for 13 days spanning over a year, with 110 time instances including more than 1000 snapshots of occupants' activities, indoor environment, working behaviour and preferences. Results showed that the shared working areas positioned near windows or in more open, connected and visible locations are significantly preferred and utilised for communication and working, and semi-enclosed space on the side with less visibility and higher privacy are preferred for focused working. The flexibility of multi-use opportunity was the most preferred feature for hybrid working. The findings offer data-driven insights for human-centric space planning and design of office spaces in the future, particularly in the context of hybrid working setups, hot-desking and co-working systems.


Acknowledgements: We would like to thank the lab_ collective, sense_ and the depot_ for enabling the data collection and sharing the valuable dataset for this study, especially to Ashish Mehra, the Tech Lead at sense_. This paper is also an extended version of ‘Occupancy level prediction based on a sensor-detected dataset in a co-working space’ presented in the 9th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys ‘22), USA.


Humans, Workplace, Privacy, Machine Learning, London

Journal Title

PLoS One

Conference Name

Journal ISSN


Volume Title



Public Library of Science (PLoS)
EPSRC (EP/T022159/1)