Soil Classification using the Stacking Ensemble Learning Technique for Crop Agronomy

Abstract

The main objective of the research work is to implement stacking ensemble learning techniques for classification of the soil types of a given region to determine the most appropriate crop to cultivate using proper irrigation systems and suitable fertilizers. Soil is a major factor in crop agronomy, and India has a various types of soil including red, black, sandy, alluvial, forest and mountain soil. The agricultural yield mainly relies on the type of soil, season (Kharif, Rabi and Zaid), irrigation method (sprinkle, surface, drip) and appropriate fertilizers. The proposed approach is being used to classify and analyze the soil of a particular region with the intent to enhance the yield of agriculture. It also helps agronomists in forecasting which crop might be preferable to cultivate and also suggesting the suitable fertilizers and irrigation systems (avoid wastage of water) to be adopted.  In this research paper, different types of soil are classified (regarding cultivation) through our proposed Stacking Ensemble Learning (classification technique) by using artificial intelligence and machine learning techniques. The resulting decision tree serve as valuable tool for farmers and agricultural practitioners to understand the optimize crop selection based on prevalent soil conditions.  The proposed method uses three base classifiers (KNN, Random forest and XGBoost) and a meta_learner (AdaBoost) to create an Ensemble model. Compared to existing works (SVM, KNN, Decision tree and Bayesian Model algorithms), the soil classification result using our proposed stacking ensemble learning approach to decision tree is more accurate.