Enhancing Stock Market Predictions with a Stacked CNN-LSTM Ensemble Model

Abstract

The prediction of stock market movements remains a complex and challenging task due to the dynamic and intricate nature of financial markets. In recent years, deep learning models have emerged as promising approaches for capturing patterns and dependencies in financial time series data. This research paper presents an innovative approach to stock market prediction using a stacked ensemble consisting of CNN-LSTM networks, for multiple tasks.

The proposed Stacked CNN-LSTM Ensemble (SCLE) model leverages the unique strengths of CNNs and LSTMs to effectively learn hierarchical representations from historical stock market data. The CNN component focuses on extracting local features and capturing short-term dependencies by operating on the raw input sequences. Meanwhile, the Bidirectional LSTM component models the long-term dependencies and temporal dynamics inherent in the sequential data, both from front to back, and vice-versa. Such is the ubiquity of this model architecture, that it shows state- of-the-art performance for both time-series prediction as well as sentiment analysis.

Training data was collected by methods of scraping news articles for sentiment analysis and using Python libraries to get the latest closing value of a particular selected stock.

To evaluate the performance of the proposed SCLE model, extensive experiments were conducted, to see its performance in capturing the historic trends. Multiple companies from the Fortune 500 list were chosen for this task. The results demonstrate that the proposed ensemble outperforms individual CNN and Bidirectional LSTM models, as well as other commonly used baseline methods for stock market prediction. The SCLE model achieves higher accuracy, robustness, and generalization capabilities, resulting in improved forecasting of stock market movements, using both historical data as well as current market sentiment analysis.

Overall, this research paper introduces a novel ensemble model, the SCLE model, which combines CNNs and Bidirectional LSTMs for stock market prediction. The proposed Ensemble model is an attempt at generalizing the predicted value, to incorporate not only temporal stock trends but also quantize current market sentiment, into a meta-model that has the potential to outperform other state-of-the-art models.