Abstract

Accurately predicting long term real estate values is a critical yet underexplored domain, Existing models focus on short term real estate value prediction that too only for developed regions. This study introduces GeoVest, a scenario aware machine learning approach to predicting long term land values in America. GeoVest (RMSE:0.053) focuses on both developed and underdeveloped regions taking macroeconomic indicators such as GDP, population, and employment rate for model training. This system integrates Random Forest with trend based forecasts to predict land values for up to 20 years in the future under three scenarios, best-case, middle range case and worst-case scenario. The framework is deployed as a user-facing mobile application, where users can select the location (per state), land size and duration of investment. This effectively presents the user with a range of values for a specific location of their choosing. While it does not consider localized factors such as infrastructure projects, zoning changes, etc, its strong baseline performance highlights the visibility of macroeconomic drivers for scalable land value prediction. Future work will incorporate localized development data to improve resilience and accuracy further. Overall, Geovest demonstrates the potential to make long term forecasts by combining ensemble learning and trend based analysis.