Soil Fertility Forecasting with Neutrosophic-Based Uncertainty Management

Abstract

This study presents a machine learning framework that combines Random Forest classification with neutrosophic logic to predict soil fertility from 16 physicochemical properties. The model achieves robust classification accuracy (95%) while introducing a novel uncertainty quantification mechanism through neutrosophic truth (T ), indeterminacy (I) and falsity (F ) values derived from prediction probabilities, assigning high-confidence (T = 0.9) to probabilities ≥0.9, indeterminacy (I = 0.3) to borderline cases (0.5≤p<0.9), and falsity (F=0.5) to low-confidence predictions (<0.5), enabling tiered decision support where 60% of cases receive automated recommendations, 25% require expert review, and 15% are flagged for rejection. Feature importance analysis reveals Clay (24.1%) and CEC (18.7%) as dominant predictors, aligning with agronomic principles while providing interpretable results. This approach advances precision agriculture by supplementing binary classifications with actionable confidence metrics, particularly valuable for managing uncertain soil conditions.