Crop Yield Prediction Using Deep Reinforcement Learning

Abstract

Predicting crop production using information about the environment, soil, water, and crops themselves is an area ripe for investigation. Deep learning-based models often extract crop attributes for prediction. These techniques have the following drawbacks: The quality of the extracted characteristics substantially affects the efficacy of those models, and there is no simple non-linear or linear mapping between raw data and crop production estimates. Deep RL guides and inspires to compensate. Deep reinforcement learning uses RL and DL to accurately forecast agricultural yields from raw data. A Deep Recurrent Q-Network model predicts agricultural yield. This model uses Q-Learning reinforcement learning and a Recurrent Neural Network deep learning algorithm. Data parameters feed the Recurrent Neural network's layers. Q-learning networks forecast agricultural yields from data. Q-values are linear mappings from Recurrent Neural Network output values. A threshold and parametric features help the reinforcement learning agent predict crop productivity. Finally, the agent is awarded a final score that takes into account the steps taken to both reduce error and improve forecast accuracy. By maintaining the original data distribution, the suggested model efficiently predicts the crop production with an accuracy of 93.7%, exceeding competing models.