This paper presents StockSense, a full-stack web-based dashboard for real-time stock market analysis that integrates Natural Language Processing (NLP)-driven news sentiment scoring with walk-forward machine learning price forecasting. The system is implemented as a Flask backend serving a dynamic dark-themed frontend. For NLP, headlines are fetched live from Google News RSS and processed by VADER and TextBlob, producing per-headline sentiment scores that are aggregated into financial, news, and combined sentiment channels. Technical analysis — including RSI, MACD, Bollinger Bands, and moving averages — is computed on historical price data retrieved via the yfinance API. A Gradient Boosting Regressor trained on 18 engineered features using a strictly chronological 80/20 train-test split performs multi-step walk-forward price forecasting of up to 30 sessions, with no look-ahead leakage. Evaluated across multiple NSE/BSE-listed stocks (2-year window), the model consistently achieves R² above 0.99, sub-1% MAPE, and direction accuracy exceeding the 50% random baseline. The dashboard also provides stock fundamentals, a rule-based recommendation engine, and a context-aware chat assistant. StockSense demonstrates how NLP and ML can be combined in a practical, deployable educational tool for retail investors.
Vinay et al. (Fri,) studied this question.