Abstract
This work explores a method for classifying peaks appearing within a data-intensive time-series. We summarize a case study from a clinical trial aimed at reducing secondhand smoke exposure via the installation of air particle monitors in households. Proper orthogonal decomposition (POD) in conjunction with a k-means clustering algorithm assigns each data peak to one of two clusters. Aversive feedback from the monitors increased the proportion of short-duration, attenuated peaks from 38.8% to 96.6%. For each cluster, a distribution of parameters from a physics-based model of airborne particles is estimated. Peaks generated from these distributions are correctly identified by POD/clustering with >60% accuracy.