Deep learning-driven soil quality prediction: integrating feature selection for enhanced accuracy
ADVANCES IN SPACE RESEARCH, cilt.77, sa.12, ss.11629-11647, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 77 Sayı: 12
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.asr.2026.03.094
- Dergi Adı: ADVANCES IN SPACE RESEARCH
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Artic & Antarctic Regions, Compendex, INSPEC, MEDLINE, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Sayfa Sayıları: ss.11629-11647
- Ondokuz Mayıs Üniversitesi Adresli: Evet
Özet
Accurate assessment of soil quality is essential for sustainable land management, yet traditional methods often involve costly and labor-intensive laboratory analyses. This study aims to develop a cost-effective and high-accuracy prediction model for the Soil Quality Index (SQI) in the semi-humid Central Black Sea region of Turkiye. Soil samples were collected from 54 locations across agricultural, forest, and pasture lands. First, a comprehensive SQI was calculated using the Analytic Hierarchy Process (AHP) based on 26 biophysical and chemical soil properties, serving as the ground-truth target. Subsequently, a 10-layer deep learning model was developed to predict these AHP-derived SQI values using soil indicators as inputs. To optimize the input dataset, meta-heuristic feature selection methods, specifically Simulated Annealing (SA) and Genetic Algorithms (GA), were employed to identify the most effective subsets of indicators. The model performance was evaluated using subsets of 25, 20, and 15 indicators. The results demonstrated that the deep learning model could predict the SQI with over 94% accuracy using a reduced subset of 20 variables, effectively eliminating redundant indicators without significant information loss. In contrast, reducing the inputs to 15 variables resulted in a notable decline in accuracy (approx. 81%). A cost analysis revealed that the 20-feature model provides the optimal balance between high explanatory power and reduced computational and laboratory costs (approx. 20% reduction). This study proposes a scalable, hybrid approach combining deep learning and meta-heuristic optimization for efficient soil quality monitoring. (c) 2026 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.