Student R&D · Applied ML

Sentiment Analysis

A research project exploring how traditional indicators, SEC filings, news sentiment, and deep learning architectures predict stock price movements.

NLP PyTorch React FastAPI
AAPL and JNJ stock price prediction charts showing LSTM, Transformer, and ARIMA model comparisons
Back to Home

The Pipeline

This project combines historical price data with hand-built technical indicators and sentiment scores extracted from SEC filings and news articles. By integrating multiple data streams—market structure, derived features, and natural language signals—we explore whether machine learning models can learn patterns that traditional econometric methods miss.

Assets & Horizon

Daily OHLCV (Open, High, Low, Close, Volume) data combined with hand-built technical indicators and sentiment scores from SEC filings and financial news sources.

Feature Engineering

  • SMA, EMA, MACD
  • RSI, Bollinger Bands
  • Sentiment scores

Models

  • ARIMA baseline
  • LSTM regressor
  • Transformer regressor

Results

LSTM Performance

The LSTM regressor demonstrated strong predictive capability, tracking both trend direction and local price swings across different equity series.

R² Score (AAPL) ≈ 0.93
R² Score (JNJ) ≈ 0.85

Key Findings

The LSTM learns to follow rallies, drawdowns, and sideways ranges where sentiment features provide additional signal that ARIMA baselines never capture. This suggests that deep learning architectures can effectively integrate heterogeneous data sources—numerical indicators and text-derived features—to improve market forecasts.