Adaptive Decision-Driven Framework for Fertiliser Optimisation Using Agroanalytics in Tropical Agriculture

ABSTRACT:Efficient fertiliser management is essential for sustainable agriculture, as imbalanced application reduces yields, raises costs and increases environmental risks. Tropical agriculture faces unique soil—crop interactions that demand region-specific solutions. However, existing studies often overlook tropical nutrient dynamics, lack robust preprocessing for noisy agricultural data, and provide limited interpretability in recommendations. To address these gaps, this study proposes an adaptive decision-driven agriculture framework that leverages soil nutrient analysis and spatial information with advanced decision-support techniques to deliver precise and interpretable fertiliser strategies. In preprocessing, an improved Density-Based Spatial Clustering of Applications with Noise algorithm is employed to detect and remove outliers by adapting density parameters to heterogeneous tropical soil data, ensuring cleaner and more consistent inputs. For soil classification, the NUTREE classifier enhances accuracy by capturing non-linear nutrient dependencies with adaptive feature weighting. Furthermore, the CSRC module with SHAP identifies key nutrient–crop relationships, translating complex outputs into human-interpretable insights. Finally, an optimised HSO-ANFIS model generates personalised recommendations, overcoming standard ANFIS limitations through faster convergence, dynamic rule optimisation and adaptability to soil–crop–region conditions. Experimental results demonstrate 97% precision and a 93.37% F1 score, with nitrogen use reduced by 33% in paddy, improving yields by 25.7% and lowering costs by 22–24%. The framework thus promotes sustainable tropical agriculture by optimising inputs, improving economic returns and reducing environmental impact.