Emitter clogging represents a significant challenge to the efficiency and sustainability of drip irrigation systems worldwide. Although machine learning holds promise for predicting clogging, existing approaches frequently exhibit limited generalizability and interpretability. To address these shortcomings, this study develops an interpretable multi-model machine learning framework. The research aims to accurately predict the emitter irrigation duration and reveal the underlying physical mechanisms. A comprehensive database comprising 450 experimental samples was constructed, covering three water types and three emitter structures. Ten key water quality and structural features were characterized as inputs. Subsequently, eight machine learning models were systematically evaluated. Among these models, XGBoost was selected as the preferred model because it achieved the best prediction stability and the lowest overall bias across cross-scenario conditions, as reflected by the lowest coefficient of variation and the prediction mean closest to 1.0. To interpret the model, SHapley Additive exPlanations were employed. This analysis quantitatively identified sediment concentration as the dominant factor negatively influencing emitter irrigation duration. Furthermore, a critical nonlinear interaction was revealed: emitter channel depth (D) exhibits a distinct regulatory threshold at approximately 0.5 mm. When D ≥ 0.5 mm, the rate of irrigation duration decline induced by increasing sediment concentration is substantially mitigated. This empirical indicator provides a useful reference for optimizing emitter geometry within the tested structural range, although further validation using more emitter geometries is required before generalization. Consequently, the proposed framework not only provides a reliable predictive tool but also delivers mechanistic insights. It thereby advances data-driven approaches from pure prediction toward informed design optimization and precision management of drip irrigation systems in complex global environments.