Integrating entropy indexing and machine learning for enhanced soil quality monitoring in organic-amended soils

• Introduction and development of an innovative Entropy-Based Soil Quality Index (ESQI) for comprehensive and cost-effective evaluation of agricultural soils. • Application of machine learning algorithms (Random Forest and Linear Regression) for accurate prediction of the soil quality index. • Identification of key soil parameters (total nitrogen, cation exchange capacity, and available phosphorus) influencing soil quality. • Demonstration of the positive effects of organic amendments (vermicompost and biochar) on improving soil physical and chemical properties. • Reduction of monitoring costs and enhancement of soil quality assessment accuracy through integrated models and selected features. This study presents an integrated framework for predicting soil quality by combining the Entropy-based Soil Quality Index (ESQI) with feature selection techniques and machine learning (ML) models. Unlike conventional SQI or PCA-based approaches, which often rely on subjective weighting, ESQI objectively quantifies the information content and uncertainty of each soil parameter using Shannon entropy, thereby enhancing interpretability, reducing input complexity, and improving predictive reliability. In this study, 21 soil samples were collected from experimental plots with varying levels of organic-amended, and 9 key physical and chemical properties—including available phosphorus (AP), total nitrogen (TN), organic carbon (OC), cation exchange capacity (CEC), anion exchange capacity (AEC), electrical conductivity (EC), bulk density (BD), porosity and pH—were measured. Advanced feature selection techniques consistently identified TN, AP, and CEC as the most influential factors affecting soil quality. ML models, including Linear Regression (LR) and Random Forest (RF), were evaluated using leave-one-out cross-validation. The LR model achieved very high accuracy when multiple features were used (R² = 0.974, RMSE = 0.023), while the RF model demonstrated excellent performance even with a single key variable (AP; R² = 0.962, RMSE = 0.019), confirming its capability to capture nonlinear relationships. Vermicompost application significantly improved soil quality. The proposed ESQI-ML framework not only provides accurate and reliable soil quality predictions but also serves as a cost-effective and practical tool for field monitoring, sustainable soil management, and data-driven decision-making in agricultural systems.