Stock Prediction with Improved Feedforward Neural Networks and Multimodal Fusion

Abstract

This paper proposes a stock prediction model based on an improved feedforward neural network, aiming to solve the shortcomings of traditional methods in processing high-dimensional nonlinear data, multimodal feature fusion, and time-dependent modeling. By introducing adaptive feature selection modules, multimodal fusion modules, and regularization strategies, the proposed model can dynamically adjust feature weights, integrate multiple modal information (such as historical prices, technical indicators, and market sentiment), and effectively suppress overfitting problems. Experimental results on multiple data sets show that the proposed model is significantly superior to mainstream methods such as ARIMA, random forest, support vector machine (SVM), and LSTM in terms of evaluation indicators such as mean square error (MSE), mean absolute error (MAE) and root mean square error (RMSE), showing higher prediction accuracy and generalization ability. In addition, ablation experiments further verify the key role of each module in improving the overall performance. Compared with traditional methods, the proposed model not only has stronger nonlinear modeling capabilities but also can capture the dynamic characteristics of the stock market, especially in multimodal data fusion. In the future, this method is expected to be expanded to scenarios such as real-time streaming data prediction, cross-market data modeling, and emergency event analysis, providing new research ideas and technical support for the field of financial forecasting.