Predicting Stock Market Movements Using News Sentiment and Machine Learning

Abstract

Stock market prediction has always been a challenging task due to the volatility, complexity, and non-linear dynamics of financial markets. Traditional approaches based on fundamental and technical analysis often fail to capture the behavioral factors that drive market fluctuations. In recent years, the integration of news sentiment analysis with machine learning (ML) techniques has emerged as a promising solution for improving predictive accuracy. By extracting sentiment scores from financial news and combining them with historical stock data and technical indicators, researchers can capture both quantitative and qualitative aspects of market behavior. This allows models to reflect not only economic fundamentals but also investor psychology, which frequently influences short-term price movements.

In this study, a hybrid approach was applied to predict stock movements of selected Nifty50 companies by combining sentiment data with ML algorithms such as K-Nearest Neighbour (KNN), Random Forest (RF), XGBoost, and Long Short-Term Memory (LSTM). The results reveal that ensemble-based models, particularly Random Forest and XGBoost, outperform deep learning and instance-based methods, achieving the lowest error rates and highest R² scores. These findings confirm that sentiment-enhanced machine learning frameworks are effective in capturing market dynamics. The study highlights the importance of integrating behavioral signals with computational intelligence, paving the way for more robust, transparent, and practical forecasting models in modern financial markets.