Ia-Mstm: Improved Adaptive Moment Estimation Enabled Mixed Attention-Based Long Short-Term Memory For Crop Yield Prediction

Crop yield prediction has sought tremendous attention in recent years several approaches were developed for crop yield prediction. However, these techniques created certain limitations in real-time applications including a reduction in relative errors, and low prediction accuracy. Therefore, this research proposed a deep learning technique known as improved Adaptive Moment estimation enabled mixed attention-based long short-term memory (IA-MSTM) to predict crop yields. The mixed attention techniques combine the advantages of channel attention with the hard attention mechanism (HAM), which can capture long-term dependencies that are crucial for understanding crop growth dynamics over time. Additionally, the improved repeated Loss-based Adam optimizer adapts the learning rate during training, leading to quicker convergence compared to traditional methods. The performance of the IA-MSTM achieves a minimal error value in terms of Mean Absolute Error (MAE) of 2.00 for TP 90.