Enhancing Stock Market Prediction with Sentiment-Augmented Random Forest

Abstract

This paper investigates stock trend prediction despite the challenges due to the numerous influencing factors and the stock market's dynamic, non-linear, and complex nature. While statistical models have laid groundwork in stock prediction, recent advances in quantitative finance emphasize intelligent timing and stock selection through machine learning. Machine learning models, particularly, have shown promise by effectively learning the relationships between predictor variables and stock movements, often outperforming traditional statistical approaches in both accuracy and robustness. This study systematically develops a stock forecasting model that combines technical indicators and sentiment analysis, employing exponential smoothing for refining technical indicators and using an optimized Random Forest model with dynamic weight adjustments and sentiment scores derived from Yahoo Finance data.
Key research area of this paper is the integration of textual sentiment analysis via the FinGPT model, a transfer learning model trained extensively on financial content, which significantly enhances sentiment-based stock prediction. The study evaluates the optimized Random Forest model’s performance in medium- and long-term forecasting, assessing its effectiveness alongside SARF, RF, and LSTM models through comparative metrics. This integration of sentiment with technical indicators aims to better capture the nuances of stock movement and the impacts of market sentiment, contributing to improved predictive accuracy.