Machine learning-enhanced multi-gas discrimination with a miniaturized MOS sensor array

Accurate and miniaturized gas sensing has become increasingly essential for real-time environmental monitoring and industrial safety. This study proposes a computationally enhanced microelectronic gas sensing platform that combines a compact metal oxide semiconductor (MOS) sensor array with machine learning algorithms for selective multi-gas detection. The core of the system is the MICS4514 sensor, which integrates two miniaturized MOS sensing elements into a single package, enabling detection of carbon monoxide, nitrogen dioxide (NO₂), ammonia (NH₃), and hydrogen (H₂) across various concentration levels. Sensor output data were processed using several supervised machine learning models, including Decision Tree, Random Forest, Quadratic Discriminant Analysis (QDA), and Gradient Boosting. While QDA yielded the highest accuracy in initial classifications, data augmentation strategies significantly improved GB's performance, achieving 100% accuracy in gas discrimination. In addition, linear regression analysis was employed to estimate gas concentrations, demonstrating its feasibility for quantitative sensing. This integration of microscale sensor technology and data-driven computational modeling underscores the potential of embedded intelligence in low-power, cost-effective gas sensors. The approach presented here supports the development of scalable on-chip sensing solutions for smart electronics and Internet-of-Things (IoT)-enabled environmental surveillance systems.

Graphical Abstract