Stock Vision: A Multivariate LSTM-Based Stock Market Analytics and Prediction Web Application

Abstract: Contemporary financial markets generate incomprehensible volumes of structured and unstructured data for each trading session, but most retail investors, small fund managers, and individual participants lack the technology to extract actionable insights from that data in real-time. This paper introduces Stock Vision, an integrated intelligence platform on stock markets that combines a five-feature multivariate Long Short-Term Memory deep learning model with a live analytics dashboard based on the Streamlit web framework. The prediction model analyzes sixty days of historical context in five engineered channels: closing price, twenty-day moving average, daily trading volume, a fourteen-period Relative Strength Index, and a news sentiment score to predict short-term price projections. The surrounding platform offers live commodity pricing, nine technical analysis charts, a portfolio tracker, market breadth statistics, and an amalgamated news feed. Results from observations demonstrate that the multivariate configuration yields more directionally accurate output than single-channel baselines by embedding the dynamics of momentum, trend regime, and event-driven sentiment at once in a concordant feature representation.

Keywords: LSTM, multivariate time series forecasting, sentiment analysis, RSI, technical indicators, deep learning, Streamlit, portfolio analytics, NSE, yfinance.