Stock Price Prediction using Technical Data and Sentiment Score

Abstract

This study provides a detailed comparison of machine learning models for stock price prediction using technical and sentiment analysis. Key findings highlight the effectiveness of LSTM with technical data and the benefits of integrating sentiment analysis from BERT with XGBoost for long-term predictions. Sentiment analysis using the modified BERT obtained an accuracy of 92.73%. On the other hand, BERT-LSTM, BERT-XGBoost, and LSTM evaluated using MAE, RMSE, MSE, and MAPE metrics also showed satisfactory results. Although LSTM performs well with technical data alone, combining BERT sentiment analysis with XGBoost shows promise for improving accuracy, especially in long-term predictions. Future research should focus on model refinement and data quality to improve predictive analysis in financial markets. Recommendations are to explore fundamental indicators and refine sentiment analysis by considering user influence.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint