A Novel Adaptive Xgboost-Based Deep Neural Network for Agricultural Yield Forecasting

Abstract

The prediction of crop yield is a very important component in smart agriculture to enable farmers and policy makers to optimize the use of resources, improve food production planning, and enhance agricultural sustainability. Yet, conventional machine learning models are unable to capture the intricate nature of the nonlinear relationships that exist between climatic, soil, and cultivation factors and are important to crop productivity. To solve this problem, this paper introduces an Adaptive XGBoost-Guided Deep Neural Network (AXG-DNN) model to improve the crop yield prediction performance. The proposed framework combines the capability of feature selection of Extreme Gradient Boosting (XGBoost) and the nonlinear learning power of Deep Neural Networks (DNN). Agriculture data with climatic, soil and crop management parameters is pre-processed and normalized first. Then XGBoost is used to assess the importance of the features and find the most significant features impacting the yield of the crops. After selecting these features, the selected features are fed to a multi-layer DNN optimized using the Adam algorithm for yield prediction. The proposed model can prune away unnecessary features and keep significant features, thus decreasing computational complexity and enhancing prediction results. The results of the experiments show that the proposed model is more accurate in prediction, robust and generalized as compared to other machine learning and deep learning models such as Multiple Linear Regression (MLR), Artificial Neural Network (ANN), Support vector regression (SVR), K-Nearest Neighbour (KNN) and Random Forest (RF). The proposed approach would enable scalable and efficient precision agriculture, enabling intelligent decision-making and sustainable crop management.