Crop Recommendation Model using Machine Learning based on Soil Analysis

Agriculture is central to international food security but it is coming under increasing pressure from climatic change, soil erosion and inefficient use of inputs. This study introduces a machine learning based crop recommendation mechanism that replaces experience-based crop selection with decision support mechanism based on data provided by key soil and weather features that include Nitrogen (N), Phosphorus (P), Potassium (K), pH, rainfall, humidity and temperature. Multiple supervised learning models are combined using stacked ensemble strategy to improve the prediction stability especially the closely related crop classes. A modular and connectivity aware input layer provides for the use of inputs for both manual data entry, and sensor based data acquisition, and for use in environments with limited or with intermittent internet connectivity. Experimental evaluation on publicly available Kaggle Crop Recommendation data set has shown as Random Forest classifier the accuracy of test dataset reaches 99.55% and ensemble model achieves 99.32% in lower variance validation coefficient which shows improvement in model robustness. Results proving that the suggested framework is effective and reproducible and well adapted to precision agriculture and to future IoTenabled smart farming systems.