Machine Learning Models for Stock Market Prediction Using Sentiment and Performance Data: A Review

Abstract

Stock market prediction has emerged as one of the most challenging and attractive applications of machine learning (ML) and deep learning (DL). The non-linear, dynamic, and highly volatile nature of financial markets makes traditional statistical models insufficient for accurate forecasting. Recent studies increasingly combine historical performance indicators (OHLCV, technical indicators, financial ratios) with sentiment data derived from news, Twitter, Reddit, and financial reports to improve predictive capability. This review comparatively analyzes machine learning models used between 2021 and 2023, focusing on supervised ML, deep learning, and hybrid sentiment-aware architectures. A synthesis of 10 recent published studies reveals that LSTM, Random Forest, XGBoost, and hybrid CNN-LSTM models outperform conventional regression methods, particularly when sentiment signals are integrated. However, major limitations persist in interpretability, overfitting, real-time adaptability, and transaction-cost-aware evaluation. The paper concludes with research gaps and future directions for robust stock forecasting systems.