A glimpse into real-world kitchens: Improving our understanding of cookstove usage through in-field photo-observations and improved cooking event detection (CookED) analytics

Abstract The combustion of solid fuels in residential cookstoves is a global health and climate issue, and expanded use of improved cookstoves could have significant benefits locally and globally. Evaluating impacts of improved cookstove programs requires more accurately measuring stove use patterns. This work builds on and improves existing stove use monitoring methods. First, we introduce and describe a novel, in-field photo-observation sampling method designed to capture near-continuous, real-world, ground-truth stove usage information. These measurements are used to validate predictions made by electronic stove use monitors (SUMs). Second, we present Cooking Event Detector (CookED), a SUM algorithm that translates stove-temperature measurements into classifications of cooking or not-cooking. The predictive performance of the new algorithm is evaluated using results from the photo-observations and compared to existing algorithms. CookED demonstrates considerable improvement over some methods for all five types of improved and traditional stoves monitored in the study. Overall minute-level predictive accuracy of CookED ranges from 95.6% to 98.4%, depending on the stove type, while Matthews correlation coefficients range from 72.8% to 88.3%. Comparisons between predicted and observed average cooking event durations show high correlation (Pearson's r = 0.85). These methods can be applied in a wide variety of applications, including research studies linking behavior, technology, exposure, and human and environmental health, as well as operational programs that aim to scale up improved cookstove adoption and quantify benefits.