Enhancing Stock Market Prediction with an LSTM-Attention Model Integrating Macroeconomic and Sentiment Features

The forecasting of stock price movements has remained an open issue for a long time because of the unpredictable and nonlinear patterns of the data. To improve the precision of time series models for stock market prediction, this paper presents a new LSTM -attention model. The proposed method is different from statistical and machine learning models as LSTM can model long-term dependencies in the temporal data and attention layer is used to handle with noise in stock price temporal data. Furthermore, this paper proposes an enhancement to feature engineering where the model has more contexts about the market through the use of macroeconomic indicators, sentiment analysis of financial news articles, and historical stock prices. Application of the model to historical data of major indices is shown in experiments that at an average of 89.7% prove the model to be a notable improvement over the baseline models. The methodology for time series forecasting is considered to be an improvement over current approaches for advancing accuracy of predictions in financial applications, especially stock markets, and is useful for investors and analysts in decision-making.