Forecasting Land Prices in Colombo District, Sri Lanka using Machine Learning and Spatial Network Analysis

Abstract

Land prices are essential indicators of economic and social development, but forecasting them presents challenges due to the multitude of factors that influence land value. This study explores the use of Machine Learning and geospatial analysis to forecast land prices in 50 key cities within the Colombo District of Sri Lanka. The dataset used comprises customer-submitted land price information provided by a real estate company ‘Lanka Property Web’ collected between 2018 and 2023. Geospatial data, including shapefiles, were obtained from the Survey Department of Sri Lanka and used to generate new variables, such as distances from land plots to significant infrastructure and locations, through spatial network analysis.A range of machine learning algorithms were evaluated for their performance in predicting land prices, including Linear Regression, Gradient Boosting, XGBoosting, Random Forest, and Artificial Neural Networks (ANN). Model performance was assessed based on predictive accuracy, with XGBoosting emerging as the most accurate, achieving an accuracy of 62.23%. In comparison, Linear Regression achieved 8.29%, Gradient Boosting 60.01%, Random Forest 58.54%, and ANN 48.01%.To further extend the forecast, a Seasonal Autoregressive Integrated Moving Average (SARIMA) model was applied to predict land prices up to the second quarter of 2027. The final deliverable of this research is a user-friendly dashboard designed for individual land buyers and sellers. This dashboard provides dynamic, data-driven insights into land market trends, empowering users with the information needed to make informed decisions in a rapidly changing property market. This study demonstrates the potential of machine learning and geospatia analysis in improving land price forecasting, offering valuable tools for stakeholders in the real estate sector.