Asbestos, due to its resistance to temperature, chemicals, and mechanical damage, was widely used in construction, especially in the second half of the 20th century. Today, most of the remaining asbestos products are roofing sheets, which pose a serious health risk and the WHO calls for its elimination. One of the key actions is to identify buildings with asbestos-cement roofing. As part of the research, a method for detecting such buildings was developed using machine learning, LiDAR technology, orthophotos, and satellite images. The developed model is based on data such as reflection intensity, roof slope, and year of construction. The proposed methodology allows for the effective localization of buildings covered with this hazardous material. The model successfully identified between 79% and 91% of asbestos-covered buildings across three diverse test areas, and up to 90% of buildings were correctly classified in terms of roofing material. Quantitative evaluation demonstrates that the proposed method effectively identifies asbestos-covered buildings achieving precision values ranging from 0.59 to 0.89 and recall between 0.77 and 0.86. The resulting F1-scores (0.70–0.88) confirm a strong balance between correct and false detections. Detection accuracy was influenced by environmental factors such as tree coverage and surface contamination, which introduced visual noise in RGB imagery and occasionally led to misclassification. Despite these limitations, the results confirm the high potential of this approach for asbestos detection. The study demonstrates that integrating LiDAR intensity, roof geometry, and visual data significantly improves the reliability of asbestos-cement roof identification compared to single-source methods.