Applications and Material Design of Machine Learning-Enabled Metal Oxide Semiconductor Gas Sensors

Because of their facile fabrication and ease of integration, MOS gas sensors have been widely used in environmental monitoring, industrial safety, and health diagnostics. However, under complex application scenarios, MOS sensors still face challenges such as cross-sensitivity, environmental interference, device drift, and strongly nonlinear response signals, which limit their detection accuracy and long-term stability. Machine learning offers significant advantages in high-dimensional data processing, nonlinear mapping, and pattern recognition, thereby providing new avenues for improving the performance of MOS gas sensors. Centered on the sensing mechanisms and data characteristics of MOS gas sensors, this review systematically summarizes recent progress in the application of machine learning to quantitative gas concentration analysis, gas species identification, drift compensation, and stability enhancement, as well as the machine-learning-assisted design of gas-sensing materials, while comparing the roles and applicability of different models in feature extraction, signal interpretation, and performance optimization. In addition, this review summarizes the challenges facing the field, including limited sample sizes, model generalization, interpretability, and edge deployment, and further outlines future research directions. This review may serve as a useful reference for the deeper integration of machine learning with MOS gas sensors and for the development of intelligent gas-sensing systems.