<p>This study proposes a novel ensemble network model called ELinear, which is based on the Linear model. This model utilizes mutual learning algorithms and channel-independent techniques to effectively reduce the error rates of Linear models in stock price prediction. We conducted comprehensive comparisons between ELinear and several state-of-the-art (SOTA) forecasting models using stock price data from companies in mainland China. The results show that ELinear achieved the lowest average prediction errors, with improvements ranging from 1.8 to 9.7% over SOTA models. Additionally, by comparing with basic models, we found that the Linear model has significant advantages in stock price prediction compared to mainstream LSTM models and newer Transformer models. Finally, based on the predictions made by ELinear, we performed validity analysis and profit assessment. The results indicate that the validity of the ELinear model’s predictions can be as high as 90%, and applying it to stock trading could achieve an average increase in returns of 9%.</p>