Improving Accuracy in Low-Cost Gas Sensors Without Machine Learning

Abstract Description: Low-cost sensors (LCS) for air quality monitoring often face accuracy challenges due to environmental factors, cross-sensitivities, and sensor drift. Traditional solutions, reliant on Machine Learning (ML) models, have limitations such as requiring extensive training datasets, frequent recalibration, and lack of adaptability to diverse environmental conditions. Recognizing these constraints, Kunak, the manufacturer of our Apex AIR multi-pollutant sensor systems, has developed novel correction techniques that bypass ML to enhance sensor accuracy under varying environmental conditions.

These corrections address the effects of high humidity and extreme temperatures, conditions under which LCS typically perform poorly. The sensors maintained high correlation in both cold and humid conditions in Sweden, where NO₂ and NO sensors achieved R² values above 0.93, and in Kuwait’s arid climate, where O₃ measurements remained accurate (MAE = 5.64 ppb) despite temperatures reaching 47°C. The evaluation under California’s AQSPEC program further demonstrated the effectiveness of the correction algorithms, as all sensor types displayed strong correlation coefficients and low MAE values. Notably, intra-model variability was minimal, with deviations below 1 ppb, ensuring consistent measurements across different units. Additionally, Kunak’s multi-pollutant system was the only one in the AQSPEC evaluation where all sensors met the EPA’s target values, highlighting its superior reliability. These findings demonstrate that air sensor performance can be consistently maintained across diverse climatic conditions for different sensor types when appropriate correction algorithms and quality control techniques are applied.

Kunak’s approach eliminates the need for external data or ML postprocessing, making it more scalable and adaptable. These advancements significantly reduce errors compared to reference instruments and enable LCS to achieve near-reference accuracy. By overcoming traditional limitations, this technology enhances the spatial and temporal resolution of air quality monitoring, complementing regulatory instruments in diverse environments.

: