Exploiting Air Quality Monitors to Perform Indoor Surveillance: Academic Setting

TL;DR

This work investigates the privacy implications of indoor air-quality data by showing that eight lab activities can be inferred from multi-sensor air pollution signals collected with the DALTON platform. The authors collect three months of annotated data in an academic lab, engineer sliding-window features across pollutants, and evaluate several lightweight classifiers, with Random Forest achieving a peak $F1$-score of $97.7\%$. The results reveal strong activity-to-signal mappings and highlight potential privacy risks associated with third-party access to indoor air data, while also outlining plans to extend the approach to additional activities and spaces. Overall, the study demonstrates the feasibility of indoor surveillance via air-quality side-channels and emphasizes the need for privacy-preserving data governance in such systems.

Abstract

Changing public perceptions and government regulations have led to the widespread use of low-cost air quality monitors in modern indoor spaces. Typically, these monitors detect air pollutants to augment the end user's understanding of her indoor environment. Studies have shown that having access to one's air quality context reinforces the user's urge to take necessary actions to improve the air over time. Thus, user's activities significantly influence the indoor air quality. Such correlation can be exploited to get hold of sensitive indoor activities from the side-channel air quality fluctuations. This study explores the odds of identifying eight indoor activities (i.e., enter, exit, fan on, fan off, AC on, AC off, gathering, eating) in a research lab with an in-house low-cost air quality monitoring platform named DALTON. Our extensive data collection and analysis over three months shows 97.7% classification accuracy in our dataset.

Figures (7)

• Figure 1: Readings from five colocated DALTON devices indicate the variability across sensors made by the same vendor. The maximum error between two devices is within the error margin, as reported by the vendor.
• Figure 2: Air monitor's measurements due to different activities - (a) Accumulation of CO2 when students enter the classroom and drop upon exit during an exam, (b) Temperature change with AC on/off, (c) VOC spike when eating food.
• Figure 3: Variation of CO2 concentration with indoor activities throughout the day. We observe that CO2 correlates with indoor occupancy or gathering.
• Figure 4: Data collection - (a) Deployment of DALTON modules in four corners of the research Lab, (b) Proportion of the recorded activity classes in the dataset. We observe that the most frequent activities are members entering and exiting the lab, where prohibited practices like eating or gathering are very infrequent.
• Figure 5: Overview of the data processing and modeling pipeline for activity classification. The features are computed across all neighboring devices by applying statistical operators. We use off-the-shelf ML models to classify indoor activities.