Short-Term Agricultural Commodity Price Forecasting: A Metaheuristic-Optimized LSTM–Attention Framework

Accurate short-term agricultural commodity price forecasting is important for market monitoring, procurement planning, inventory management, and price–risk awareness. This study develops a data-driven forecasting framework, namely CMACS–ARO–LSTM–Attention (CALA), which combines an LSTM–Attention predictor with an improved Artificial Rabbits Optimization algorithm. The original reported CALA results are retained unchanged. This minimal revision adds persistence/random-walk and ARIMA references on the archived comparison blocks. For wheat, the reported CALA result is R2=0.995, RMSE =0.008, MAE =0.004, and MAPE =0.021%; the newly added persistence/random-walk and ARIMA references have RMSE values of 0.102 and 0.092, respectively. The soybean application is presented as supplementary mixed evidence rather than proof of broad generalization. The high goodness of fit is interpreted as short-horizon price-level approximation in a persistent series, not as proof of directional trading profitability. Repeated CALA runs, formal predictive-accuracy tests, equal-budget ablations, and multi-step interval forecasts remain outside this minimal revision.