To investigate the damage evolution mechanism of fractured sandstone under freeze-thaw cycles, an integrated research framework combining numerical simulation, machine learning, and decision logic analysis & interaction effect analysis was established. Numerical models were constructed using PFC software, through which freeze-thaw cycle tests and uniaxial compression simulations were conducted. A deconstruction formulation was developed to isolate the contributions of freeze-thaw damage and pre-existing fissures. The XGBoost algorithm was employed to build a multi-objective prediction model achieving a test set R² exceeding 0.97, while the SHAP method was applied to interpret decision logic and variable interactions. Results indicate a transition from brittle to ductile failure modes under freeze-thaw action, with mechanical property deterioration rates gradually decelerating and showing damage accumulation saturation. The micro-macro chain damage evolution mechanism is summarized as frost heaving forces driving particle bond degradation, triggering crack propagation and force chain network failure. The cumulative effect of this progressive damage manifests macroscopically as a significant reduction in load-bearing capacity. SHAP analysis further reveals three interaction effects on compressive strength: synergistic deterioration between freeze-thaw cycles and upper temperature/fissure length/fissure thickness; synergistic compensation between cycle count and high dip angles (>45°); and antagonistic effects between cycles and fissure quantity. Finally, an integrated XGBoost-SHAP platform was developed, providing an intelligent tool for rock mass stability assessment in cold regions.