A Systematic Review of Explainable AI Methods and their Applications in Geotechnical Engineering

Abstract

In recent years, artificial intelligence (AI) has achieved remarkable results in various fields; however, the opacity of its decision-making process limits its deployment in safety-critical domains and undermines user trust. This paper presents a conceptual review of the latest research progress in Explainable Artificial Intelligence (XAI). This review constructs a conceptual classification framework to categorize mainstream explanation methods into two types: pre-hocinterpretability (characterized by inherent transparency) and post-hoc interpretability (relying on ex-post analysis). The latter is further divided into “perturbation-based” and “backpropagation-based” paradigms. Secondly, the paper delves into the core contradictions of different methods in terms of theoretical completeness, computational efficiency, and explanation validity. Then, it focuses on innovative application cases of XAI in the field of geotechnical engineering to verify its technical adaptability in complex real-world scenarios. Finally, from an interdisciplinary perspective, we propose key directions for future research, providing theoretical support and practical approaches for building trustworthy AI systems.