An Intelligent Module for Estimating the Investment Attractiveness of Agricultural Lands in a Region

Abstract

This research presents an intelligent module for estimating the investment attractiveness of agricultural lands, addressing the complex interaction of geospatial and economic factors in Russia. We developed a novel hybrid methodology combining Gower distance-based clustering with SHAP (SHapley Additive exPlanations) feature weighting to generate an interpretable investment index. Applied to a dataset of 335 land objects characterized by eight mixed-type features, our approach automatically determined an optimal two-cluster structure, effectively distinguishing high-potential objects not far from the infrastructure from remote, less attractive objects. Crucially, we derived data-driven feature weights via a surrogate Random Forest model trained on cluster pseudo-labels, overcoming the limitations of traditional linear hedonic models. The resulting system achieved exceptional performance, demonstrating 98.5% classification accuracy and an F1-score of 0.97. Furthermore, the regression component for rental rate adjustment yielded a low Root Mean Square Error (RMSE) of 15.87 rubles. Implemented in Python 3.12, this framework significantly enhances decision-making for land portfolio management, enabling precise, fair market valuation and optimized rental strategies for agricultural investors. Hence, the developed method can significantly simplify the management of large real agricultural objects portfolios, providing a transparent and reasonable pricing strategy.