This study presents a novel approach for real-time gas identification at room temperature. We use UV-modulated Sb-doped SnO2 sensors combined with machine learning. Our method exclusively employs the gas response (R) as the sole metric. This eliminates the need for time-dependent parameters such as response and recovery times. By modulating the UV light intensity at five distinct levels (5, 10, 15, 20, and 30 mW/cm2), we generate a five-dimensional optical fingerprint. This fingerprint captures subtle variations in sensor response under different illumination conditions. Gas discrimination was evaluated for both oxidizing gases (O3 and NO2) and reducing gases (NH3 and H2). Our machine learning results show that Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) achieve nearly 100% accuracy when four UV intensity levels are used. Using R as the sole input metric allows for instantaneous response detection, which is essential for real-time gas identification. This approach addresses the limitations of conventional thermally activated sensors that require multiple parameters and paves the way for the development of rapid-response monitoring systems.
Keywords: Sb-doped SnO2; UV-modulated; machine learning; optical fingerprint; selectivity.