Routledge, H; Tang, R; Stamp, S; (2025) Advancing indoor environmental monitoring: A machine learning approach for post-data collection activity validation. In: Journal of Physics : Conference Series. Institute of Physics (IoP): Lausanne, Switzerland.

Preview

Text Routledge_2025_J._Phys.__Conf._Ser._3140_112018.pdf Download (2MB) | Preview

Abstract

Occupant behaviour is a significant driver of air quality in indoor spaces, but due to the innate variability of these activities, they can be complex sources. This makes it difficult to assess their direct health impact. Validation methods currently used to research occupant activities are also often disruptive and vulnerable to inaccuracies, making research in this area difficult. This pilot study aims to establish whether it is possible to develop a new method to conduct activity research by identifying activities post-data collection. Household environmental data was collected in 3 houses over 3 months, alongside participants completing an activity diary. This data was used to identify an activity’s ‘signature’ and develop generalised and individual activity identification models using machine learning (ML). The generalised activity model achieved 55-64% overall accuracy when tested on training data, and approximately 30% when applied to non-training data. These results indicated that a single model was not the best approach to accurately identify events. Using individual specialised models increased accuracy significantly, achieving 72-92% when applied to the training dataset, and 57-79% when applied to the non-training dataset. This suggests that the method is viable, and it is possible to accurately identify various activities using household environmental data.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item