Machine learning-enhanced metal oxide gas sensor arrays: a review of selectivity and performance improvements

Purpose This study aims to critically examine the integration of machine learning (ML) techniques into metal oxide (MOX) gas sensor arrays to address the challenges of poor selectivity and long-term drift. It analyzes the architecture of intelligent sensing systems, explores advanced data preprocessing and feature engineering techniques and evaluates a wide range of machine learning (ML) algorithms for gas classification and drift compensation. Design/methodology/approach A comprehensive literature survey was conducted to identify peer-reviewed studies involving MOX-based electronic nose (e-nose) systems enhanced by ML algorithms. Key areas explored include sensor system architecture, dynamic sensing modes, feature extraction and selection methodologies, and classification models ranging from support vector machines and ensemble methods to deep learning architectures. Selection followed explicit inclusion criteria (time-/batch-split evaluations, MOX arrays, quantitative outcomes), with emphasis on 2023–2025 studies comparing one- and two-dimensional convolutional neural networks pipelines and recent drift-adaptation methods. Findings ML techniques have significantly improved the selectivity and accuracy of MOX gas sensors. Ensemble and boosting methods, such as XGBoost, offer high performance with minimal training time, outperforming some deep learning models in structured data scenarios. Temperature-cycled operation coupled with feature-rich time-series data enhances gas discrimination. Continuous learning frameworks show exceptional long-term stability, maintaining performance over multiple years despite sensor drift. However, the field still faces unresolved challenges, including reliable detection in complex gas mixtures, lack of standardized benchmarks for performance comparison and the need for explainable AI in safety-critical applications. Originality/value This review provides a detailed synthesis of current advancements in ML-assisted MOX sensor arrays for gas sensing, highlighting the shift from materials-driven selectivity to data-driven intelligence. By evaluating the entire intelligent sensing pipeline – from sensor response acquisition to model deployment – it offers valuable insights for researchers, engineers and practitioners developing next-generation gas sensing systems. It also outlines a future research roadmap emphasizing codesign of materials and algorithms, multimodal data fusion and real-world validation.