Predicting Stock Trends Using Web Semantics and Feature Fusion

Abstract

Stock data are characterized by high dimensionality and sparsity, making stock trend prediction highly challenging. Although the Light Gradient Boosting Machine (LightGBM), based on web semantics, excels at capturing global features and efficiently performs in stock trend prediction, it does not consider the issue of declining prediction performance caused by the changing distribution of stock data over time (concept drift phenomenon). Accordingly, this work introduces the Convolutional Neural Network (CNN) into the prediction model to leverage its ability to effectively capture local features. Additionally, local features are combined with global features to obtain a comprehensive set of feature information. Lastly, the model processes new data in real-time, continuously learns new knowledge, updates model parameters, and effectively addresses the decline in model performance caused by concept drift. Experimental results demonstrate that the proposed model outperforms other models indicating its ability to efficiently perform well in stock trend prediction.