📈 SentimentEdge — Stock Sentiment & Price Movement Analyzer

Does market sentiment predict stock price direction?
An end-to-end NLP + ML pipeline using live Yahoo Finance data — no API keys required.

---

🖥️ Dashboard Preview

Overview — Price × Sentiment

90-day candlestick chart with live sentiment overlay, KPI metrics, and bullish/bearish signal

Heatmap — Sentiment Distribution & Cross-Company Comparison

Sentiment score distribution, daily post volume trends, and multi-stock sentiment comparison across AAPL, AMZN, GME, UBER, NFLX

ML Predictor — Model Leaderboard & Next-Day Signal

Next-day directional prediction with probability scores, walk-forward CV leaderboard (Accuracy, Precision, Recall, F1, ROC-AUC), and confusion matrix

Correlation — Lag Analysis

Pearson correlation across 0–7 day lags, OLS scatter plot, and key findings panel — identifies exactly when sentiment spikes reach the market

---

🎯 What It Does

SentimentEdge answers a real financial question: can news sentiment and analyst signals predict whether a stock will go up or down tomorrow?

1. Collects live news headlines, analyst upgrades/downgrades, and trading volume from Yahoo Finance
2. Scores each item using VADER NLP sentiment analysis with volume-weighted aggregation
3. Engineers 40+ features including RSI, MACD, Bollinger Bands, ATR, OBV, and sentiment-price interaction terms
4. Trains three ML classifiers evaluated with walk-forward (TimeSeriesSplit) cross-validation — no data leakage
5. Visualises everything across 5 interactive dashboard tabs

---

🚀 Quick Start

# 1. Clone the repo
git clone https://github.com/prithvinairr/sentimentedge.git
cd sentimentedge

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the dashboard
streamlit run app.py

# 4. Select a company in the sidebar → click RUN ANALYSIS

No API keys needed. All data pulled live from Yahoo Finance.

---

📊 Dashboard Tabs

Tab	What It Shows
Overview	Candlestick price chart with sentiment overlay, rolling correlation, KPI metrics
Heatmap	Sentiment score distribution, daily post volume, cross-company comparison
ML Predictor	Next-day signal, probability scores, model leaderboard, confusion matrix, feature importance
Correlation	Lag analysis (0–7 days), Pearson r, OLS scatter with trendline
Data Explorer	Raw tables for all collected data, CSV export

---

🤖 Machine Learning

Models Trained

Model	CV Accuracy	Precision	Recall	F1	ROC-AUC
Logistic Regression	58.4%	61.5%	54.6%	54.7%	0.640
Random Forest	55.3%	60.7%	60.5%	56.3%	0.578
Gradient Boosting	55.8%	58.4%	53.1%	53.4%	0.568

Baseline (random guessing) = 50%. All models trained with 5-fold TimeSeriesSplit — trains strictly on past data, tests on future. Zero data leakage.

Feature Engineering (40+ Features)

Category	Features
Sentiment	Lags 1–5d, 3/7/14-day MA, trend, acceleration
Technical	RSI(7,14), MACD, Bollinger Bands %, ATR, OBV
Price	MA crossovers, momentum, volatility
Interactions	Sentiment × RSI, Sentiment × MACD, Sentiment × Volume

---

🏗️ Architecture

Yahoo Finance (yfinance)
        │
        ├── OHLCV Prices ──────────────────────────────┐
        │                                              │
        ├── News Headlines ──→ VADER Sentiment ────────┤
        │                                              │
        └── Analyst Ratings ──→ Sentiment Score ───────┤
                                                       │
                                          SQLite Database (5 tables)
                                                       │
                                        Feature Engineering (40+ features)
                                                       │
                                     ML Training (TimeSeriesSplit CV)
                                                       │
                                       Streamlit Dashboard (5 tabs)

---

📁 Project Structure

sentimentedge/
├── app.py                      # Streamlit dashboard entry point
├── pipeline.py                 # Data pipeline orchestrator
├── database.py                 # SQLAlchemy schema (5 tables)
├── requirements.txt
├── assets/                     # Screenshots for README
├── utils/
│   ├── yfinance_scraper.py     # Live news & analyst data collection
│   ├── stock_fetcher.py        # OHLCV price downloader
│   └── sentiment_aggregator.py # Daily score aggregation
└── models/
    └── predictor.py            # Feature engineering + ML training

---

💡 Key Findings

• Sentiment shows peak correlation at Lag 0 (Pearson r = +0.362 for AAPL) — the market reacts same-day to sentiment spikes
• Positive sentiment → price rises direction confirmed across multiple tickers
• ML models consistently outperform the 50% random baseline across all 5 CV folds
• Cross-company sentiment comparison reveals sector-level sentiment clustering

---

📄 License

MIT — free to use, fork, and extend.