<p>This study introduces a novel machine learning (ML) framework for daily soil moisture prediction across India, integrating Multivariate Empirical Mode Decomposition (MEMD) with advanced regression techniques. The proposed hybrid model consists of three main stages: (i) MEMD-based decomposition of multiple meteorological and hydrological inputs—including surface runoff, evaporation, total precipitation, surface pressure, and solar radiation; (ii) prediction of each intrinsic mode function (IMF) component using various ML algorithms, namely Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), and Support Vector Regression (SVR); and (iii) reconstruction of the final soil moisture time series from the predicted components. Input data were sourced from ERA5, the fifth-generation ECMWF reanalysis dataset, which provides high-resolution atmospheric and hydrological variables capturing the spatiotemporal variability of soil moisture drivers from 2015 to 2023. Results indicated that the hybrid RF-MEMD model consistently outperformed its counterparts across different locations and input combinations showing R² values of 0.76–0.95 and RMSE values of 0.18–3.47. Moreover, the model’s predictive accuracy was noticeably lower in arid and semi-arid regions with complex fluctuations than in wet areas, regardless of the ML method or hybrid variant used. The application of MEMD enhanced the modeling framework by enabling the simultaneous decomposition of multivariate signals with aligned IMFs, thus improving the representation and forecasting of complex nonlinear soil moisture dynamics.</p>