Chemiresistive sensor array for quantitative prediction of CO and NO2 gas concentrations in their mixture using machine learning algorithms

• Published in: Mikrochimica acta (1966) Vol. 191; no. 12; p. 756
• Main Authors: Naganaboina, Venkata Ramesh, Jana, Soumya, Singh, Shiv Govind
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Analytical Chemistry; Binary mixtures; Carbon monoxide; Characterization and Evaluation of Materials; Chemistry; Chemistry and Materials Science; Classification; Cobalt sulfide; Environmental monitoring; Gases; Graphene; Machine learning; Microengineering; Nanochemistry; Nanotechnology; Nitrogen dioxide; Real time; Room temperature; Sensor arrays; Sensors; Target detection; Zinc oxide; Chemiresistive sensor array; Air pollutants; Classification and regression; Selectivity and cross-sensitivity; Low-cost gas sensors; Machine learning
• ISSN: 0026-3672; 1436-5073; 1436-5073
• DOI: 10.1007/s00604-024-06835-x

Single sensors have been developed for specific gas detection in real-time environments, but their selectivity is limited by interference from other gases when considering mixtures of gases. Consequently, accurate detection of target gases in mixed gas environments is essential. Therefore, this study develops a sensor array approach to quantitatively estimate the concentration of carbon monoxide (CO) and nitrogen dioxide (NO 2 ) gases in their binary mixture (CO and NO 2 ). The sensor array consists of two different sensors, developed with zinc oxide and graphene-cobalt sulfide. The sensor array was tested in the presence of 29 different proportions of the binary mixture at room temperature. Subsequently, machine learning (ML) algorithms are applied on sensor signals to estimate the concentration of gases. The ML models unfortunately exhibited inaccurate prediction when all sensor signals were considered, therefore, to improve the prediction accuracy, the sensor signals were divided into three levels based on the mixed gas concentration regime. Interestingly, the classification and regression algorithms provided good classification accuracy (85.13 ± 3.2%) and reasonable predictions of target gas concentrations at three levels. The proposed computational framework can be extended to include additional gases in mixed gases and used in various applications, including automotive, industrial, and environmental monitoring. Graphical abstract