Forecasting Financial Futures: A LSTM Framework for Stock Market Prediction

This work provides a transparent, end-to-end hybrid framework for stock price forecasting with the use of a Long Short-Term Memory (LSTM) neural network. Our network is structured to identify and understand old patterns by mixing two different data sources: firstly, quantitative technical indicators and secondary qualitative news sentiment analysis. The paper explains a comprehensive preprocessing setup, which takes into account the obtaining of 3 years of historical data, the production of daily sentiment scores, Min Max Scaler normalization for all features, and the generation of 60-day look-back sequences. Details about the model's working concepts as well as the method of its training, which is enhanced by Dropout and Early Stopping shows in this paper. The model's prediction accuracy is measured by R-squared and Root Mean Squared Error. The main point of this article is that it is an easily understandable, reproducible, and “humanized” guide to mixing technical and sentiment data.