Answers are grounded in this paper and cite their sources.
Although gas sensing technologies are widely applied in environmental monitoring, industrial safety, and food testing, their applications remain constrained by several challenges, including cross-sensitivity, long-term drift, environmental interference, and slow response. Machine learning (ML) provides new opportunities for intelligent gas sensing by extracting useful patterns from complex sensing signals and improving gas recognition and quantitative analysis. Adopting a problem-oriented perspective, this review examines recent progress of machine learning technology in gas sensing and detection. First, ML methods used in gas sensing are introduced according to learning paradigms and sensing tasks. The applications of these methods in solving key gas-sensing problems are then discussed. Specifically, extracting multichannel and dynamic response features improves gas discrimination; domain adaptation and active learning can compensate for gas drift, whereas including temperature and humidity signals can reduce environmental interference; early-response analysis enables faster gas detection without waiting for steady-state responses. This review further discusses the challenges that limit the practical deployment of ML-based gas sensing systems. Potential solutions, transferable from intelligent sensing applications in other fields, are introduced to solve data scarcity, edge deployment difficulty, and poor interpretability problems, such as data-efficient learning, lightweight model design, and interpretable modeling. Finally, future perspectives such as biomimetic olfactory systems, embodied intelligence for gas source tracking, and AI-assisted analysis of new gas sensing materials are discussed. This review provides an overview of the recent progress of machine learning technology in gas sensing, highlighting future opportunities for developing reliable and automatic gas detection systems.
No takes yet. Share an insight, caveat, or question.
Pan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: