Stock price sequences are nonlinear, non-stationary, and noise-contaminated. This paper proposes a hybrid RVMD-ConvNeXt-BiLSTM-ECA model that tackles two recurring weaknesses in existing frameworks: insufficient local feature extraction and high computational cost. A RIME-optimised VMD (RVMD) module decomposes the raw price series into stationary intrinsic mode functions (IMFs), with mode number K and penalty factor α selected automatically by minimising sample entropy. A ConvNeXt block then extracts local coupling features from the IMFs and filtered technical indicators. A two-layer BiLSTM captures temporal dependencies, followed by an efficient channel attention (ECA) module that re-weights feature channels with minimal overhead. A two-layer fully connected head produces the final forecast. Experiments on four A-share stocks from the Tushare financial database show that the model improves both prediction accuracy and inference speed over strong baselines.

Hailong Sun, Adisak Sangsongfa, Noppadol Amdee. Stock Price Trend Forecasting via a Hybrid RVMD-ConvNeXt-BiLSTM-ECA Framework. Academic Journal of Computing & Information Science (2026), Vol. 9, Issue 4: 61-69. https://doi.org/10.25236/AJCIS.2026.090408.

[1] DIXON M. Industrial forecasting with exponentially smoothed recurrent neural networks. Technometrics, 2022, 64: 114–124.

[2] GULMEZ B. Stock price prediction with optimised deep LSTM network with artificial rabbits optimisation algorithm. Expert Systems with Applications, 2023, 227: 120346.

[3] LIU H, ZHAO T, WANG S. A stock rank prediction method combining industry attributes and price data of stocks. Information Processing & Management, 2023, 60(4): 103358.

[4] WANG Q, WU B, ZHU P, et al. ECA-Net: efficient channel attention for deep convolutional neural networks. Proceedings of CVPR, 2020: 11531–11539.

[5] SU H, ZHAO D, LIU L, et al. RIME: A physics-based optimisation. Neurocomputing, 2023, 532: 183–214.