Optimized hybrid deep learning model for banana price prediction

Agricultural product price prediction is significant for agricultural economic management and market decision-making. In this article, taking banana price forecasting as a representative case study, a hybrid prediction framework that integrates Variational Mode Decomposition (VMD), Particle Swarm Optimization (PSO) and Deep Learning models is proposed. The framework improves prediction accuracy through synergistic integration of signal decomposition, parameter optimization, and ensemble learning. Firstly, PSO is used to optimize the modal parameters of VMD to decompose the price series adaptively. Secondly, we develop a hybrid model combining Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN), and Transformer architectures, termed the LSTM-TCN-Transformer Hybrid Model, to respectively capture long-term dependencies, local temporal features, and global contextual patterns in the time series data. Finally, the multi-model prediction results are integrated by Random Forest. Experiments demonstrate that this model outperforms single models and traditional ensemble methods in terms of root mean square error (RMSE) = 0.0079, mean absolute error (MAE) = 0.0061, mean absolute percentage error (MAPE) = 0.11%, and coefficient of determination (R 2 ) = 0.9991, validating its efficiency and robustness.