Predicting Stock Price Movements Using Daily News

• Tech Stack: Python, LSTM, Natural Language Processing, BERT, Time-series Analysis, HPC, GPU, Bayesian Optimization
• Github URL: Project Link

This project compares different approaches to forecasting next-day stock price movements. It begins with a baseline model that uses LSTM networks with historical stock data as an input. Subsequently, Bidirectional Encoder Representations for Transformer (BERT) classification model is constructed. The final model combines LSTM network and BERT to predict stock prices by extracting embeddings from daily news headlines. Our model demonstrates improvement in the final model’s predictive accuracy, with Mean Average Precision increasing by 0.11 to 0.56 and the AUC score rising by 0.16 to 0.59 compared to the baseline model. This highlights the value of integrating advanced machine learning techniques with news headlines for stock market forecasting.