After the 2015 Nepal earthquake, thousands of homes had to be rebuilt. Tracking this recovery usually requires in-person surveys, which can be limited and costly. In this study, we use satellite radar data, known as InSAR, to remotely measure how much rebuilding has occurred. InSAR detects changes on the Earth’s surface, but the data can be noisy, especially in Nepal’s complex landscapes and weather conditions. To address this, we apply machine learning models that help separate actual reconstruction activity from background noise. We tested several of these models using data from over 8,500 households to find the most effective approach. Our findings show that regional recovery can be monitored using satellite technology, helping to fill gaps left by limited ground surveys. This method can support disaster response teams by highlighting areas that still need assistance and could eventually be used in real-time recovery planning in other locations.