3TFL-XLnet-CP: A Novel Transformer-Based Crop Yield Prediction Framework with Weighted Loss Based 3-Tier Feature Learning Model

The advancement of crop yield prediction through artificial intelligence (AI) has gained significant attention. However, the existing AI-based approaches for maximizing agricultural productivity, specifically in crop yield prediction, have not consistently delivered satisfactory results. In response to this challenge, we propose a novel framework named as Three Tier Feature Learning with XLnet based Crop Prediction (3TFL-XLnet-CP) that enhances agricultural productivity by accurately predicting crop yield. The 3TFL-XLnet-CP framework employs a three-tier feature learning approach in combination with the powerful XLnet transformer-based crop prediction model. The three-tier feature learning involves the integration of Spiking Neural Network (SNN), Graphical Neural Network (GNN), and Convolutional Neural Network (CNN) to extract distinct feature vectors from the preprocessed data. These feature vectors are then concatenated using Jaccard Similarity to measure their similarity score. Additionally, a weighted Loss function is introduced to optimize feature learning, further enhanced by a novel self-adaptive Spider Monkey Optimization algorithm (SASMO). The concatenated features are subsequently fed into the classification layer for making precise crop yield predictions. The proposed model is implemented using the Python platform and evaluated against existing models such as ANN, RNN, DNN, and BiLSTM. The comparison demonstrates the superiority of our proposed 3TFL-XLnet-CP framework in accurately predicting crop yield.