The Application of Random Forest Prediction in Developing a Systematic Land Parcel Value in the Urban Area

Land administration services, especially in urban areas, require a complete and accurate parcel-based land valuation for supporting fair and reliable taxation. For fulfilling that purpose, the utilization of machine learning techniques, instead of full ground survey, to predict land value for each parcel is tested in this work. Although linear regression approach is widely used, the technique is not relevant as not all variables are linearly related to land values. Over the past few years, random forest (RF) models have been applied and seen to be promising for land-administration related analysis. However, as the prediction was typically generating land values for zones or blocks, which are not parcel-based prediction, thus the produced results may not be operational for local land and municipality offices to be used especially for determining land values in property transactions. This study adopted RF to predict parcel-wise land values in 15 administrative districts of Surabaya City and are validated using real transaction data. The independent variables used in this study are as follows: the zoning score, road width, parcels’ distance from the central business districts (CBD), schools, markets, arterial roads, urban collector roads, points of interest (POI), and hospitals. The prediction model was trained using assessment values done by Indonesian Society of Land Appraisers. The prediction model produces the mean absolute error (MAE), mean absolute percentage error (MAPE), R²-score, Coefficient of Variation (COV), and Prediction relative difference (PRD) respectively on the testing data are 39.31 USD.; 6.99%; 0.96; 4.18%; and 1.03. Then, the trained model was validated using official transactional dataset and determined its MAPE and R²-score. The values were 16.76% and 0.72, respectively. It is concluded that the model performed effectively. Hence, the findings provide a potential solution for delivering completeness and certainty in land valuation for land registration services.