Overview
Project Kassandra is a stock price prediction pipeline that combines traditional technical analysis with multi-source sentiment data. The system fetches live market data, aggregates sentiment signals from news, Wikipedia, and Google Trends, and uses machine learning to predict next-day closing prices.
Features
- Live Data Fetching: Historical stock prices via yfinance
- Technical Indicators: Moving averages, volatility, daily returns
- Multi-Source Sentiment:
- News sentiment (Google News RSS + VADER)
- Wikipedia pageview trends
- Google Trends search interest
- Explainable Fusion: Fixed-weight sentiment aggregation (0.4 news, 0.3 trends, 0.3 wiki)
- Rolling Predictions: Day-by-day training with no future data leakage
- CSV Artifacts: Exportable features and prediction logs
Tech Stack
- Python 3.8+
- pandas, numpy
- scikit-learn (RandomForestRegressor)
- feedparser, nltk (news sentiment)
Future Plans
- Payment gateway expansion.
- Analytics dashboard.
- Microservices migration.
Limitations & Future Work
Current Limitations:
- News sentiment limited to recent articles (Google News RSS)
- Google Trends data may have rate limits
- Model uses fixed hyperparameters (no tuning)
- Single-day prediction horizon
Future Enhancements:
- Hyperparameter optimization
- Multi-day forecasting
- Additional sentiment sources (Twitter, Reddit)
- Deep learning models
- Real-time prediction API
