Natural disasters like earthquakes can kill a lot of people and damage a lot of buildings. We need a good evacuation plan to cut down on those numbers. Traditiona systems frequently fail when confronted with unpredictable disaster scenarios due to their dependence on human decision-making and predetermined strategies. This paper delineates an earthquake evacuation system that is predicated on machine learning, with the objective of enhancing route planning and identifying safe areas. It trains models such as Random Forest, Decision Tree, K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) by utilizing geographic data such as latitude and longitude and historical disaster records. These models are beneficial for the classification of potentially hazardous regions and the recommendation of less risky areas for evacuation. The interface, which is based on Python Tkinter, allows users to input coordinates, train models, and upload data. Subsequently, predictions are generated in real time. Folium can also be employed to emphasize disaster areas in various colors to enhance the readability of maps. The Random Forest model dominates the competition in terms of accuracy, response time, and decision-making.