A predictive analytics model with Bayesian-Optimized Ensemble Decision Trees for enhanced crop recommendation

Researchers have been working on developing new effective, reliable, and environmentally friendly crop recommendation systems. This study introduces a thorough framework for predicting crop recommendations by combining sophisticated machine learning (ML) classifiers with a multi-class classification approach. The study is intended to (1) comprehensively assess the importance of environmental alterations and soil nutrient characteristics across a variety of crop classes, (2) develop effective predictive analytics models using the fine Gaussian support vector machine (FGSVM) and coarse k-nearest neighbors (Coa-KNN) algorithms, (3) reduce the dimension of feature vectors and minimize training time (FGPCASVM-CRP) approach through principal component analysis (PCA), (4) explore and analyze a Bayesian optimized ensemble decision tree for crop recommendation prediction (BOEDT-CRP) model based on assessment specifications, and (5) compare the proposed approach with multiple ML classifiers with various hyperparameter optimization, including FGSVM, coarse Gaussian SVM (Coa-GSVM), wide neural network (WNN), trilayered neural network (TNN), Fine k-nearest neighbors (FKNN), cosine k-nearest neighbors (Cos-KNN), bagged tree ensemble (BTE), and subspace discriminant ensemble (SDE). The proposed models throughout the training and testing stages reveal outstanding results, with comparable accuracy rates of 99.5%, precision rates of 99.49% and 99.55%, recall rates of 99.49% and 98.59%, and f1-scores of 99.5% and 99.54%. The findings support the conclusion that the proposed models can significantly support farmers in intelligent crop management and harvesting, leading to enhanced production and decreased reliance on human labor. • Propose a novel Bayesian-optimized ensemble decision tree approach. • Develop a predictive model for crop recommendation based on soil and environmental data. • Combine PCA and machine learning to reduce feature dimensionality and training time. • Predict crop recommendations using a multi-class classification methodology.