Immittance-based machine learning framework for breath biomarker detection using Pd-In2O3-decorated MXene sensors

초록

Conventional gas sensors based on high resistivity materials often suffer from poor signal stability and limited reliability under direct current (DC) operation. To overcome these limitations, this study introduces a frequency-dependent immittance-based sensing strategy integrated with deep learning for the selective and stable detection of acetone and ethanol, even under humid conditions. The metallically conductive MXene was transformed into a high resistivity composite via thermal oxidation and decoration with Pd-loaded In2O3 nanoparticles (5 mol% relative to In2O3), enabling reliable alternating current (AC) mode operation. Complex immittance variables-including impedance, admittance, capacitance, and modulus-were extracted and employed as both numerical and image-formatted inputs for neural network models. Notably, one-dimensional convolutional neural networks (1D CNNs) trained on image-formatted impedance data significantly outperformed conventional deep neural networks (DNNs), achieving accurate gas classification even with single-frequency input. This work highlights the strong synergy between high-frequency AC sensing and image-formatted deep learning, establishing a robust and scalable platform for2real-world breath nalysis using high-resistivity two-dimensional nanomaterials.

키워드