Year: 2025
Type: conference-paper
Abstract: Deep learning methods have greatly enhanced time series forecasting in recent years. Although, forecasting agricultural prices remains particularly difficult due to their nonlinearity, volatile fluctuations, numerous influencing factors, and notable seasonal patterns. To solve these problems, this paper puts forward a new hybrid framework, TCN-LSTM-MLP-Attention, which integrates Temporal Convolutional Networks (TCN) for capturing multi-scale dependencies, LSTM for identifying long-range trends, MLP for nonlinear feature transformations, and an attention mechanism that emphasizes critical time steps. Experiments on the Hass avocado dataset indicate that the proposed model decreases MAE by 15.3% relative to a stand-alone LSTM and surpasses CNN-LSTM by 10.7%, while also achieving superior scores in MSE, RMSE, and R2.These gains underscore its utility in agricultural price forecasting and supply chain management.
Authors: Xiao Chen, Punyaphol Horata, Khamron Sunat
Institution: Khon Kaen University
Language: English
Field: Decision Sciences
Domain: Social Sciences
SDG: Zero hunger
Open Access status: gold
OpenAlex ID: W4411888400