Assessment of the Neutrosophic Fuzzy-AHP and Predictive Power of Some Machine Learning Approaches for Maize Silage Soil Quality

Abstract

Agricultural soil quality has become a noteworthy subject for study because of the rising awareness surrounding sustainable farming, the decrease in farmland due to urban growth and population increase, and the harmful environmental consequences. Hence, this study aims to use the Neutrosophic Fuzzy-AHP (NF-AHP) and Standard Scoring Function (SSF) approaches to detect the Soil Quality Index (SQI) and evaluate the accuracy of machine learning models such as Multiple Linear Regression (MLR) and Random Forest Regression (RFR) in predicting the SQI for maize silage at Yesil Kure agricultural farmland, situated in the Black Sea region of T & uuml;rkiye. To collect soil samples, the study area was divided into 300 m x 300 m grid squares, and 89 grid points were sampled (0-20 cm depth). To specify the SQI, the 28 soil quality parameters were classified into four groups: physical indicators (sand, clay, silt, bulk density, hydraulic conductivity, aggregate stability, wilting point, field capacity, available water content and slope), chemical indicators (pH, electrical conductivity, organic matter and cation exchange capacity), productivity indicators (total nitrogen, phosphorus, potassium, calcium, magnesium, sodium, iron, copper, manganese and zinc), and biological indicators (soil respiration, metabolic quotient and microbial biomass carbon). Statistical analyses as descriptive statistics, p-values, importance of covariates, geo-statistics and 10-fold cross-validation also were performed for this study. According to the NF-AHP process, slope (0.0746), OM (0.0595), TN (0.0327), and MBC (0.0787) were determined to have the highest weighting values for physical, chemical, productivity, and biological soil indicators, respectively. Also, the results indicate that MLR outperformed RFR, exhibiting considerably lower error indices (MAE: 0.005, MSE: 0.000, RMSE: 0.000) and a higher R-2 value (0.99) than RFR (0.82) through 10-fold cross-validation. The present study enhances our comprehension of soil quality dynamics and provides stakeholders with better predictive tools for SQI. The effectiveness of the MLR model in this context highlights its usefulness in agricultural decision-making and land management. In summary, our study showcases the MLR model's capacity to elevate SQI predictions. This knowledge equips practitioners with enhanced tools for sustainable land stewardship. Further refinement can be achieved through the exploration of diverse models and expansion of the scope of investigated factors, while recognizing the potential unknowns that lie ahead.