Akademik Veri Yönetim Sistemi
Computers and Electronics in Agriculture, cilt.229, 2025 (SCI-Expanded)
Agricultural lands play a pivotal role in sustaining human existence, serving as primary sites for food production, contributing to the economy, fostering biodiversity, maintaining environmental equilibrium, and upholding social and cultural values. This research investigated the suitability of agricultural areas in the Büyük Menderes River Basin (BMB), the largest basin in the Aegean Region of Türkiye. To achieve the greatest agricultural yield and ensure the most efficient utilisation of soil, an Agricultural Land Suitability Analysis (ALSA) was conducted, integrating parametric classification techniques such as Multi-Criteria Decision Making (MCDM) with nonparametric methods such as Machine Learning (ML). The objective was to devise a hybrid model that amalgamated both parametric and nonparametric approaches. The utilisation of the Analytic Hierarchy Process (AHP), Fuzzy Decision Making Trial and Evaluation Laboratory (DEMATEL), and Simple Multi-Attribute Rating Technique (SIMOS) techniques alongside Random Forest (RF), Logistic Regression (LR), and Artificial Neural Networks-Multilayer Perceptron (MLP) algorithms enabled a comprehensive evaluation of 17 criteria affecting BMB agricultural lands to be achieved. Geographical Information System (GIS) mapping facilitated the visualisation of the obtained data. The weight assignment to the criteria was accomplished via MCDM techniques, with the final determination being made by Borda ranking. Subsequently, ML models were trained within delineated areas of the basin and then applied comprehensively. Resultant ALSA maps, categorized into four classes (very suitable, moderately suitable, slightly suitable, not suitable), delineated land suitability. Parametric classification techniques revealed that moderately suitable areas were the most prevalent, covering 58.39% of BMB, while very suitable areas constituted a minimal 0.33%. Conversely, non-parametric techniques depicted larger portions of unsuitable areas, with RF (59.7%), LR (67.4%), and MLP (77.0%) highlighting this aspect. It is notable that there is considerable variation in the proportions of areas across the algorithms. The performance of the algorithms was evaluated using the f1-score metric, which demonstrated that MLP, LR, and RF achieved 99%, 98%, and 97%, respectively, thereby confirming their effectiveness. The synthesis of diverse classification methodologies facilitated the identification of optimal techniques for ALSA, thereby alleviating pressure on agricultural lands that share similar characteristics. This research highlights the value of adopting hybrid approaches in agricultural land assessment, providing actionable insights for the sustainable management and preservation of land resources.