Determining whether a wireless receiver is located indoors or outdoors is a fundamental capability for emerging spectrum-sharing systems that rely on environment-aware power control and interference mitigation. Existing approaches based on GPS or Wi-Fi often suffer from limited indoor availability, infrastructure dependence, or coarse environment characterization. In this paper, we explore the use of widely available aircraft Automatic Dependent Surveillance–Broadcast (ADS-B) signals as a signal of opportunity for passive indoor–outdoor detection. Building on a prior ADS-B–based framework, we systematically study how model selection and training strategy affect classification performance in a direction-aware setting. We evaluate three tree-based learning models—Random Forest, XGBoost, and LightGBM—under multiple data aggregation strategies, including file-level training, message-count slicing, and time-based slicing. Our optimized LightGBM-based pipeline achieves up to 94% accuracy with consistently low false-negative rates across 41 diverse datasets and deployment environments. These results demonstrate that ADS-B reception dynamics encode robust environmental signatures, enabling reliable and low-cost indoor–outdoor detection suitable for spectrum-aware wireless systems.