<p>Radon (²²²Rn) is a globally recognized Class A carcinogen, and its accumulation in urban areas presents a critical challenge for public health and spatial planning. Effective environmental management requires accurate and scalable risk assessment, especially in geologically complex cities. The study presents the first systematic assessment of soil gas ²²²Rn in Yerevan, Armenia, combined with multi-depth profiling of natural radionuclides (²²⁶Ra, ²³²Th, ⁴⁰K). This study addresses this issue by integrating a robust, data-driven Principal Component Regression (PCR) predictive framework to generate a hazard map of soil gas ²²²Rn activity across the urban environment of Yerevan, Armenia, and reveal key environmental and geological factors influencing soil gas ²²²Rn. The model integrates ²²²Rn activity measurements (ranging from 483.0 to 38,375.0 Bq/m³) with a comprehensive dataset of key predictor variables: multi-depth natural radionuclide activity concentrations, soil texture properties, and meteorological parameters, collected across a stratified sampling network. The resulting PCR prediction model with three component explains 33.6% of the variance in log-transformed ²²²Rn. Predictive power is primarily driven by PC1 (gamma-emitting radionuclide abundance and fine-grained soil texture) and PC2 (measurement depth and coarse-textured soils). Leave-One-Out Cross-Validation (LOOCV) confirmed structural model stability (cross-validated R² = 0.115 on log scale), although extreme values were conservatively underestimated. The resulting hazard map delineates radon-prone zones primarily in central-eastern and southern districts associated with permeable sedimentary formations. Despite moderate explanatory power, the PCR framework provides an interpretable and statistically robust basis for preliminary radon hazard zoning in geologically heterogeneous urban areas.</p>