AGRICULTURAL LAND PRICE PREDICTION USING MACHINE LEARNING ALGORITHMS

Abstract

The paper considers the prediction of land value using LightGBM, Fast Tree, and Fast Forest machine learning methods. The training dataset is collected from the Internet and contains of 800 rows (in some cases there are data gaps for certain attributes). An overview of the problems of collecting such data is made. From the full dataset (D1), three additional ones were created: without data on land rent price (D2), without data on normative monetary assessment (NMA) (D3) and without data on rent price and NMA (D4). Using the LightGBM method has the best prediction results, but in some cases Fast Tree’s prediction quality is similar to LightGBM. Removing NMA data from the dataset improves the quality of prediction, because after calculations it turned out that there is no correlation between market value and NMA. Also, unlike real estate value prediction, gradient boosting methods provides better results. A correct prediction is defined as a prediction with no more than 10% error, which follow similar approaches in other researches. Depending on the dataset and the chosen method of machine learning, the quality of prediction ranges from 35% to 92% of correct predictions. The paper describes research’s limitations and possible ways to improve the quality of land price predictions for reviewed machine learning methods.