Prediction Land Value Using Geographically Weighted Extreme Learning Machine

With the advancement of technology, there is an expectation that technology can simplify and accelerate the land valuation process, especially by utilizing machine learning methods to predict land value. This research explores the Geographically Weighted Extreme Learning Machine (GWELM) approach to predicting land values. GWELM combines the Extreme Learning Machine (ELM) method with geographical weighting from Geographically Weighted Regression (GWR). The evaluation results show that the GWELM model has better prediction accuracy compared to the ELM model without geographic weighting. From the evaluation results, the GWELM model shows a decrease in Mean Absolute Error (MAE) by 2.89%, a decrease in Mean Absolute Percentage Error (MAPE) by 1.49%, and an increase in the coefficient of determination (R2) by 11.48% compared to the ELM model. With the combination of ELM and GWR methods, land value prediction becomes more accurate, although it requires higher computational costs. This research uses geospatial data stored in Shapefile (SHP) format, one of the most commonly used geospatial data formats in Geographic Information Systems (GIS). To process and extract the necessary information from this data, specialized tools such as ArcGIS are required, which allow the extraction of coordinates and other features. This research shows how the GWELM approach can improve the land value prediction process by considering the geographical aspects of the data.