An Integrated Machine Learning Framework for Precision Agriculture: Disease Detection, Yield Prediction, Smart Irrigation, and Soil Classification

Abstract

Feeding almost 10 billion people by 2050 requires a 70% rise in agricultural output, while the usable land area is dwindling, water sources are increasingly under strain, and changing climate conditions continue to upset cultivation practices. Traditional farming techniques, based as they are on observation and the same application of resources, are not designed to handle such complexities and require smart and information-driven decision support systems that will allow farmers to plan for their crops and ensure adequate output amid such uncertainties. In this paper, we have developed an integrated ML model that tackles four issues related to precision agriculture, namely: (1) automatic crop disease detection, (2) multiple variable crop yield forecasting, (3) classification of soil fertility, and (4) irrigation scheduling. The proposed model uses transfer learning with the EfficientNet-B3 CNN model for detecting crop diseases among 14 plant species belonging to 38 classes. Yield forecasting makes use of a stacked ensemble consisting of Random Forest, XGBoost, and Gradient Boosting regressor models, trained using 23 input variables. Irrigation optimisation is achieved through a two-layer LSTM network capable of modelling sequential processes in soil moisture and evapotranspiration. The soil fertility assessment component is realised using an SVM classifier with an RBF kernel, supplemented with SHAP (SHapley Additive exPlanations) for interpretable predictions. On publicly available agricultural benchmark datasets, we obtain: 96.4% accuracy for plant disease detection (F1 score = 0.963), an R2 of 0.914 and RMSE of 3.21 q/ha for yield prediction, a 32.1% decrease in seasonal irrigation amount, and 91.2% soil type identification accuracy. Besides technological effectiveness, the system is explicitly aligned with United Nations Sustainable Development Goals SDG 2, SDG 13, and SDG 15, highlighting its potential for affordable, large-scale implementation among smallholder farmers in developing countries.