📈 Stock Buy / Sell / Hold Signal Predictor

A machine learning web app that predicts a Buy / Hold / Sell rating for any stock ticker, using real historical price data (via yfinance), technical-indicator feature engineering, a scikit-learn Random Forest Classifier, and a Streamlit UI with interactive Plotly charts.

Built with the tech stack (Python + pandas/numpy + scikit-learn RandomForest + Streamlit), adapted from price regression to Buy/Hold/Sell classification.

🧠 Overview

Given a stock ticker, the app pulls recent price history, computes technical indicators (moving averages, RSI, MACD, Bollinger Bands, momentum, volatility, volume trends), and feeds them into a trained Random Forest model that predicts whether the stock's next-5-trading-day outlook looks like a Buy, Hold, or Sell, along with a confidence score, an interactive candlestick chart, and a multi-stock watchlist scanner.

📁 Project Structure

stock-signal-predictor/
├── app.py                 # Streamlit application (3 tabs: analysis, watchlist, model insights)
├── train_model.py         # Downloads data, engineers features, trains & saves the model
├── features.py            # Shared technical indicator + labeling logic (train & app both use this)
├── config.py               # Ticker universe, thresholds, lookback settings
├── requirements.txt
├── model/                  # created by train_model.py
│   ├── signal_model.pkl        # trained RandomForestClassifier
│   ├── scaler.pkl               # StandardScaler for numeric features
│   ├── feature_columns.pkl      # exact feature order expected by the model
│   └── metrics.pkl              # accuracy / classification report / confusion matrix
└── README.md

📊 Data & Labels

Data comes live from Yahoo Finance via yfinance — no static CSV needed, so the model always trains on up-to-date history. For each trading day, we look HORIZON_DAYS (default 5) days into the future:

Future 5-day return	Label
> +2%	Buy
between -2% and +2%	Hold
< -2%	Sell

⚙️ How It Works

1. Data collection — train_model.py downloads ~3 years of OHLCV history for ~25 stocks (mix of US and NSE-listed) via yfinance.
2. Feature engineering (features.py) — computes SMA/EMA ratios, RSI(14), MACD + signal + histogram, Bollinger %B and band width, 10-day momentum and volatility, and volume-change/volume-ratio — 14 features total, all normalized so the model generalizes across stocks at very different price levels.
3. Labeling — each row is labeled Buy/Hold/Sell based on the actual forward 5-day return (see table above).
4. Training — a RandomForestClassifier (300 trees, max depth 10, class-balanced) is trained on an 80/20 stratified split.
5. App (app.py) — loads the saved artifacts and exposes:
   • Stock Analysis tab — candlestick chart with SMA/Bollinger overlays, MACD and RSI subplots, volume chart, and the model's live Buy/Hold/Sell prediction with confidence.
   • Watchlist Suggestions tab — scans a list of tickers (default watchlist or your own comma-separated list) and ranks them by signal, so you get "many other suggestions" at a glance.
   • Model Insights tab — test accuracy, classification report, confusion matrix, and a feature-importance chart.

🛠️ Tech Stack

• Python 3.9+
• yfinance — live historical stock data
• scikit-learn — Random Forest Classifier, preprocessing
• pandas / numpy — data handling & feature engineering
• Streamlit — web UI
• Plotly — interactive candlestick/indicator charts