A cotton futures price prediction method integrating feature synergistic selection with temporal deep learning

Accurately predicting cotton futures prices is challenging due to difficulties in fusing multi-source heterogeneous data with inefficiency in feature extraction． To address this, a multimodal data fusion framework driven by a feature co-selection mechanism and a bidirectional long short-term memory (BLSTM) network is proposed． This framework integrates multi-source information, including futures market indicators, remote sensing image features, and investor sentiment derived from textual data． The feature co-selection mechanism facilitates hierarchical dimensionality reduction, while BLSTM captures nonlinear temporal dependencies． Our experiments validate the effectiveness of multi-source data fusion in elucidating complex drivers of price fluctuations． Our method significantly enhances prediction accuracy, reducing the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and normalized root mean square error (NRMSE) by 49.11%, 56.16%, 11.21%, and 14.47%, respectively, compared to a BLSTM model trained on non-integrated data． Feature analysis reveals that historical futures prices, market sentiment indices, and remote sensing vegetation features all contribute to the prediction． The present work offers novel insights regarding analysis of agricultural financial derivatives, also provides empirical reference for applications of multimodal data in financial modeling．