Understanding the relative roles of tectonic and climatic controls on large bedrock landslides remains a key challenge in active orogenic belts. In this study, remote sensing, field observations, and interpretable machine learning tools are integrated to investigate the factors governing large, deep-seated landslides in the Hunza region. A field-validated landslide inventory, derived from anomalous topographic and geomorphic signatures, was used as a training mask. An ALOS PALSAR DEM and multi-source geospatial data including; GNSS-derived crustal velocities, slope, aspect, curvature, elevation, drainage proximity, fault proximity, seismic hazard, rainfall, topographic wetness index, and vegetation cover. This study quantitatively assesses the relative contributions of multiple conditioning and triggering factors controlling large bedrock landslides using interpretable machine learning models. Three models (Random Forest, XGBoost, and MLP neural network) were trained and interpreted using SHapley Additive exPlanations (SHAP) to quantify feature contributions. All models achieved high predictive performance (AUC up to 0.98). SHAP-based interpretation reveals that tectonic parameters, particularly crustal velocity and proximity to active thrust faults, consistently rank as the best predictors of landslide occurrence, whereas rainfall likely acts as a secondary or transient trigger. Furthermore, spatial clustering of deep-seated landslides (e.g., Mayoon, Miacher, and Budalas) within zones of elevated deformation (~25 mm/yr) within the active collision zone is consistent with progressive rock mass weakening and long-term slope destabilization. These results highlight the importance of tectonic forcing in preconditioning large bedrock landslides, with climatic inputs facilitating episodic reactivation. The integration of geodetic data with interpretable machine learning provides a quantitative framework for improving landslide hazard assessment in such tectonically active mountainous regime.