Assessing the Neutrosophic Fuzzy-AHP Based Soil Quality Index for Sugar Beet: A Comparative Study of Multi-Class Logistic Regression, Random Forest, and One-Against Support Vector Machine Models

Abstract

Recent years have seen a marked increase in the number of studies being conducted on the quality of agricultural soil. This is attributable to three key factors. Firstly, there is growing recognition of the importance of sustainable agriculture. Secondly, there is a decreasing availability of agricultural lands due to urbanisation and population growth. Thirdly, there are negative environmental consequences resulting from these factors. The primary objective of this study was to assess the soil quality index (SQI) for sugar beet cultivation under semi-arid ecological conditions. To this end, the Neutrosophic Fuzzy-AHP and Standard Scoring Function (SSF) methods were employed. Furthermore, the performance of the multi-class logistic regression (multi-class LR), random forest (RF), and one-against-all-support vector machine (OAA-SVM) models was evaluated in predicting both the linear and non-linear SQIs. Moreover, the SQI assessment entailed the calculation of spectral vegetation indices, including the Normalised Difference Vegetation Index (NDVI) and the Red Edge-Optimised Soil Adjusted Vegetation Index (RE-OSAVI). A statistical comparison was then conducted between the results and the data obtained from the high-resolution Sentinel-2A image dated 19th July 2021. This analysis utilised the NDVI and RE-OSAVI vegetation indices within the designated research area. The findings indicated that NDVI, with an r(2) value of 0.634, yielded the most favourable outcomes for the cultivation of sugar beets when considering the non-linear SQI results (p < 0.001). The prediction models in the present study were evaluated using four metrics as Matthews correlation coefficient (MCC), accuracy rate, recall, precision, and F1-score. Furthermore, Receiver Operating Characteristic (ROC) and confusion matrix were also applied for evaluation purposes. The results indicate that RF outperformed both multi-class LR and OAA-SVM in predicting both the linear and the non-linear SQI for sugar beet cultivation, demonstrating significantly higher values on MCC (0.97, 0.99), accuracy rate (0.98, 0.99), recall (0.99, 0.98), precision (0.99, 0.98), and F1-score (0.99, 0.98). These results significantly enhance our understanding of soil quality dynamics. The efficacy of the multi-class RF model in enhancing SQI predictions is of considerable significance for the informed decision-making process in agricultural and soil management practice.