Abstract: Machine learning (ML) provides a new data-driven paradigm for geological cognition, construction optimization, and operational safety in complex and high-risk tunnel scenarios. Based on 1 633 articles from the Web of Science Core Collection, this study systematically analyzes the research hotspots and evolutionary trends of ML in tunnel engineering through scientometric analysis. Focusing on the full life-cycle chain of surrounding rock identification-tunneling optimization-health monitoring, it reviews the ML algorithms and application status in existing research. In surrounding rock identification, integrated learning and deep learning methods that incorporate geological, drilling parameter, image, spectral, and geophysical exploration data markedly enhance lithology classification and anomaly detection. For tunneling optimization, temporal and multimodal models are applied to predict advance rate and energy consumption while enabling adaptive parameter control, thereby improving the characterization of nonlinear coupling effects. In health monitoring, lining defect detection is evolving toward higher accuracy, real-time performance, and lightweight implementation. Subsequently, the primary challenges in this field are systematically categorized and summarized. Finally, future research directions are proposed across eleven aspects: data standardization ecosystem, open data platforms, multimodal data fusion frameworks, model interpretability enhancement, uncertainty quantification, deeper application of large language models, lightweight deployment, model optimization, robustness enhancement, interdisciplinary collaboration, and sustainable development. These efforts are expected to facilitate the transition of ML in tunnel engineering from theoretical exploration to practical engineering application.

Key words: tunnel engineering, machine learning, scientific metrology analysis, surrounding rock identification-tunneling optimization-health monitoring, data standardization ecosystem