Hybrid TCN-MLP-Attention Network with Bidirectional LSTM for Agricultural Price Prediction

Abstract

Agricultural market price prediction plays an essential role in supporting supply chain management, market planning, and economic decision-making. However, traditional statistical and machine learning models often struggle to capture complex nonlinear relationships and long-term temporal dependencies present in agricultural price data. To address these limitations, this study proposes an enhanced hybrid deep learning architecture that integrates a Temporal Convolutional Network (TCN), Multilayer Perceptron (MLP), Attention mechanism, and Bidirectional Long Short-Term Memory (BiLSTM) network for time-series price prediction. The TCN module captures multi-scale temporal dependencies using causal and dilated convolutions, while the MLP module performs nonlinear feature transformation. The Attention mechanism highlights important temporal features that significantly influence price trends. Furthermore, the incorporation of the Bidirectional LSTM enables the model to learn contextual information from both past and future sequences, improving feature representation and forecasting accuracy. The model is evaluated using the “U.S. Avocado Market (2015–2023)” dataset obtained from Kaggle. Experimental results demonstrate that the proposed hybrid architecture significantly outperforms conventional models such as SVM, XGBoost, and standard LSTM. The enhanced model achieves an improved R² score of 0.9748 with reduced RMSE, confirming its effectiveness and robustness for agricultural timeseries price prediction.