<p>Liquefaction susceptibility of soil is usually assessed using simple empirical methods. In this study, machine learning (ML) techniques are used for capturing complex nonlinear interactions among soil properties and building a reliable classification framework. A field standard penetration test (SPT) database from Taiwan, collected during the 1999 Chi-Chi earthquake, was used. A series of ML-based ensemble algorithms have been utilized to classify the liquefaction instances. The novelty of the study lies in the utilization of explainable artificial intelligence (XAI) techniques to provide inferential and predictive insights into feature engineering for liquefaction susceptibility prediction. The results indicated that boosting-based ensemble methods, such as gradient boosting (GB), categorical boosting (CatBoost), and histogram-based gradient boosting (HGB), provided very efficient classification. The XAI techniques Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) identified influential parameters and enhanced model interpretability for field applications. Identification of the groundwater level (<i>GWL</i>) and peak ground acceleration (<i>α</i>ₘₐₓ) as the primary factors influencing liquefaction susceptibility highlights their strong correlation with the underlying physical mechanisms driving the phenomenon.</p>