Last Updated 2/3/2026
This report establishes data cutoff thresholds for fume hood proximity sensors to categorize user presence to make data interpretation more user-friendly. Utilizing proximity data from Cambridge, Stanford, and the University of Pennsylvania—reported as the fraction of time movement is detected within 5-minute intervals—we employed K-means clustering to evaluate both 3-point and 5-point scaling models. The analysis supports the adoption of a 5-point scale ("Very Unlikely" to "Very Likely") over the 3-point alternative, as it offers superior granularity for identifying "Mixed" usage patterns critical for determining underuse. We established final weighted cutoff values at 0.057, 0.182, 0.373, and 0.741. These thresholds were validated against industry norms, with the combined active usage categories accounting for approximately 4% of the time, aligning with the standard benchmark of one hour of daily fume hood usage.
Category |
Final Cutoff Value |
Expected Distribution |
Very Unlikely |
< 0.057 |
~91.2% |
Unlikely |
0.057 – 0.182 |
~4.9% |
Mixed |
0.182 – 0.373 |
~2.1% |
Likely |
0.373 – 0.741 |
~1.1% |
Very Likely |
> 0.741 |
~0.7% |
Full Report: https://docs.google.com/document/d/1D53h3mJcqG5sgx1LWuy4yTn2lVyURDJA095KQQkJtF8/edit?usp=sharing