Machine Learning and Explainable AI for Slope Stability Assessment: A Systematic Review of Methods, Parameters, and Research Gaps

Machine learning (ML) is increasingly used to predict the factor of safety (FoS) and stability state of slopes, often combined with explainable artificial intelligence (XAI). This PRISMA 2020 systematic review searched OpenAlex, Semantic Scholar and Crossref (2000–2026), screened 6,550 unique records, mapped 1,009 eligible studies at abstract level and synthesised 137 open-access full texts. Publications grew sharply after 2020, and artificial neural networks are giving way to tree ensembles, deep learning and metaheuristic hybrids. Reported performance is high (median test 2 0.97; median accuracy 0.93) but rests mainly on synthetic limit-equilibrium or finite-element data (51 % of studies) or on small, overlapping compilations of published case histories (median 190 cases). Only 11 % of studies addressed data quality or duplicates, 39 % used cross-validation, 4 % validated on independent cases and 15 % shared code or data. Class thresholds ranged from FoS 1.0 to 1.3. About half of the studies reported explanations, mostly feature importance, sensitivity analysis or SHAP, but none validated them quantitatively against limit-equilibrium sensitivities. Priorities are curated public datasets, leakage-aware and external validation, explicit target definitions, uncertainty quantification and mechanics-based validation of explanations.