An integrated dual-interpretable framework for nonstationary hydrostatic depth forecasting in residential zones under river–groundwater interactions

• Dual interpretable ML models enhance HD forecasting and reveal key feature impacts. • River discharge and its decomposed signals are key drivers of model performance. • Extreme hydrostatic depths show distinct patterns requiring special modeling attention. • Short-term lags from BRF and decomposition improve residential HD forecasting. Low hydrostatic depth (HD) values and their temporal fluctuations in residential zones pose challenges to implementing infrastructure projects such as sewerage and drainage systems. Nonstationary interactions between river flow and groundwater further complicate HD modeling. This study proposes a dual-interpretable hybrid framework integrating machine learning models (KRR, CBR, ABR, HGBR), the Boruta Random Forest (BRF) algorithm for feature selection, and Time–Varying Filtering (TVF) for signal decomposition. The framework was applied in two residential zones, Agh-Ghabr and Saqar-Yelqi, and benchmarked against single-stage models. Results show that short-term lags significantly improve HD forecasting; BRF optimizes inputs by reducing redundancy; and TVF effectively captures the nonstationary behavior of HD and river discharge, with mid- and high-level components being most influential. The BRF–TVF–KRR configuration outperformed single-stage models, as confirmed by residual analysis showing reduced variance and increased central tendency. River discharge and its decomposed components emerged as key predictors, consistent with physical hydrological processes in the study areas. Probabilistic analysis using Weibull and Beta distributions revealed that model rankings differ in critical HD ranges, underscoring the need for separate critical value analyses. The dual-interpretable design not only enhances modeling transparency but also provides a robust decision-support tool for identifying critical low-depth thresholds, thereby offering actionable insights to mitigate risks to foundations and drainage systems in residential areas influenced by river–groundwater interactions.