A Comparative Study of Sentiment-Driven Intelligent Systems using FinBERT

We highlight the effectiveness of FinBERT in capturing financial sentiment, the superiority of ML models over traditional time-series approaches, and the importance of sentiment-driven feature integration. This paper uses Apple stock data and its associated news from Alpha Vantage and perform a comparative study on stock price prediction using SARIMA and three FinBERT-augmented ML models (SVR, RFR, and GBR). The experimental results demonstrated that FinBERT+GBR achieved the best performance (MAE: 2.5004, MSE: 10.4263, RMSE: 3.2290, R²: 0.9864), significantly outperforming SARIMA and other ML models. The results underscore the potential of sentiment-augmented ML models for practical applications, though challenges in interpretability and computational efficiency warrant further exploration.