Structural breaks and reduced ground station coverage have previously limited soil moisture (SM) analyses and validation approaches over Eastern Africa. Due to its high dependence on rain-fed agriculture, it is vital to analyze SM variability and establish reliable monitoring frameworks for the region. Despite improvements in satellite data availability, comprehensive regional analyses and product comparison remain sparse. Therefore, this paper compares the four European Space Agency Climate Change Initiative Soil Moisture (ESA CCI SM) datasets (active: scatterometer; passive: radiometer; combined: sensor fusion; gap-filled: interpolation) from 1991-2023 and develops predictive Random Forest (RF) machine learning (ML) models using climate and environmental variables. The models are spatially cross-validated, as well as evaluated against 45 International Soil Moisture Network (ISMN) in-situ stations across Kenya, Uganda, Rwanda, and Sudan. Spatial cross-validation reveals that the active and combined models perform best (R 2 : 0.57/0.59), with vegetation optical depth (VOD), precipitation, and seasonality being the most important predictors. Validation to ISMN displays moderate positive correlations and small errors for all models (R: 0.59-0.65, RMSE: 0.07-0.1 m³/m³), with the highest correlations recorded for subsurface measurements (R: 0.6-0.67, 10-20 cm below surface). This demonstrates the suitability of ESA CCI SM products for the region and expands model utility beyond surface SM monitoring. Future model applications include climate scenario studies or seasonality and predictor analyses.