Kumamoto prefecture, Japan, was hit by enormous (Magnitude 6.5, 7.3) earthquakes on 14th and 16th of April, 2016. As a result of the shocks, more than 10,000 buildings were severely damaged, and over 100,000 people had to evacuate away from their homes. After the earthquake, it took the authorities several days to grasp the locations where people were evacuating, which delayed the distribution of supply and rescue teams. This situation was made even harder since some people evacuated to places that were not designated as evacuation shelters. Conventional methods for grasping evacuation hotspots require field surveys, which take time and are also difficult to execute right after the hazard in the confusion. We propose a framework to efficiently estimate the evacuation hotspots using location data collected from mobile phones. To validate our framework, we estimated the locations that were congested with evacuees after the Kumamoto earthquake using GPS data collected by Yahoo! Japan. We also verified that our estimation results were very high, by checking the features located in each grid with high anomaly value. Moreover, for one of the non-designated evacuation hotspots, we accurately estimated the population transition of before and after the earthquake, which we validated using newspaper reports.