Synergistic Integration of Frequency-Dependent Impedance and Machine Learning in Semiconductor Metal Oxide-Based Breath Sensors for High-Performance Gas Discrimination

Frequency-dependent impedance spectroscopy in combination with machine learning offers a powerful strategy for discriminating among gas species using mutually interacting semiconductor metal oxide (SMO) gas sensors. In this study, 0.3 at% platinum-loaded SnO 2 sensing materials were employed to breath-based disease detection, with a focus on machine learning-assisted discrimination of mixtures of acetone (0.5–2.5 ppm) and ethanol (0.5–2.5 ppm) under both dry and humid environments (80% relative humidity). Data features derived from the real, imaginary, and magnitude components of complex impedance obtained at the frequency range from 10 5 to 10 4 Hz were used to enhance gas discrimination performance through supervised deep learning neural networks (DNNs). Even with a single sensor designed through structural and compositional modifications, frequency-dependent impedance features enabled accurate identification of acetone concentrations in acetone–ethanol mixtures under humid conditions, achieving 99% accuracy using single-frequency impedance data (i.e., 10 5 Hz), compared to 66% with DC-based (voltage) signals. This innovative strategy offers an effective and scalable solution for detecting not only breath acetone but also gas mixtures composed of chemically similar gas species.