Data Science Approach To Intraday Stock Prediction Using LSTM Networks And Social Media Sentiment

Because of its high volatility, short decision cycles, and non-linear behavioral influence, intraday price forecasting continues to be one of the most difficult problems in financial analytics. Traditional market prediction systems ignore the psychological and socio-behavioral factors that surface online in favor of historical time-series indicators. In order to predict short-term stock price movements at intraday resolution, this study proposes an improved Data Science framework that combines social media sentiment analysis and Long Short-Term Memory (LSTM) neural networks. FinBERT and VADER are used to combine sentiment polarity signals that are taken from public Twitter discourse and financial forums with historical OHLCV features and engineered technical indicators. Compared to baseline LSTM models trained exclusively on price data, the model exhibits a 13–18% improvement in directional accuracy. The findings support the theory that psychological cues influenced by sentiment have a big impact on intraday trading decisions. The study lays the groundwork for risk-aware decision support, real-time algorithmic trading, and hybrid AI-powered investment tools.