Deep Reinforcement Learning with Modified Reward Function for Crop Yield Prediction

Nowadays, crop yield prediction is one of the major fields in a research area, the details about the preferable crops will be useful for the farmers to raise the crops. Models based on deep learning are often utilised to extract important crop features for prediction. Despite the possibility that these approaches could address the yield prediction problem, they have the following shortcomings: unable to produce a linear mapping between crop yield values and raw data; also, the effectiveness of those models is largely dependent on the calibre of the features that were extracted. For the aforementioned drawbacks, deep reinforcement learning offers guidance and inspiration. In order to predict agricultural productivity in time series data, the suggested work builds a Deep Reinforcement Learning with the Modified Reward Function (DRL-MRF). To help the crop productivity prediction, the reinforcement learning agent combines a set of parametric features with the threshold. This model can gather more rewards to carry out the search process and maybe produce useful follow-up proposals and classification scores by using the updated reward function. The evaluated result represents that the proposed DRL-MRF method delivers evaluation metrics such as R2, MAE, MSE, and RMSE values of about 0.99, 0.81, 0.72 and 0.79 respectively which ensures the accurate yield prediction compared with existing methods.