Stock Price Predictor

Time series stock price forecasting API powered by Facebook Prophet, served via FastAPI with Docker support.

How It Works

Prophet models time series as an additive combination of components:

y(t) = trend(t) + seasonality(t) + holidays(t) + error(t)

The pipeline:

Yahoo Finance ──▶ Prophet Training ──▶ Model Serialization ──▶ FastAPI Serving
(historical data)   (fit trend +         (joblib)              (REST predictions)
                     seasonality)

1. Data Ingestion — Downloads historical adjusted closing prices from Yahoo Finance via yfinance
2. Model Training — Fits a Prophet model capturing trend and seasonal patterns
3. Serialization — Saves the trained model using joblib for fast loading
4. Serving — FastAPI endpoint loads the model and generates forecasts on demand

API Reference

POST /predict

Request:

{
  "ticker": "AAPL",
  "days": 7
}

Response:

{
  "ticker": "AAPL",
  "days": 7,
  "forecast": {
    "03/06/2024": 171.42,
    "03/07/2024": 171.58,
    "03/08/2024": 171.73,
    "03/09/2024": 171.89,
    "03/10/2024": 172.05,
    "03/11/2024": 172.20,
    "03/12/2024": 172.36
  }
}

Sample Predictions

The plots/ directory contains forecast visualizations for AAPL, MSFT, and GOOG showing predicted trends with component decomposition.

Tech Stack

Component	Technology
Forecasting	Facebook Prophet
Data Source	Yahoo Finance (yfinance)
API Framework	FastAPI + Uvicorn
Serialization	joblib
Containerization	Docker

Quick Start

Docker

docker build -t stock-predictor .
docker run -p 8000:8000 stock-predictor

Local

pip install -r requirements.txt

# Train a model
python src/model.py --ticker AAPL --days 30

# Start the API server
cd src && uvicorn main:app --reload --host 0.0.0.0 --port 8000

Interactive docs: http://localhost:8000/docs

Project Structure

stock-predictor/
├── src/
│   ├── main.py         # FastAPI application
│   ├── model.py        # Prophet training and prediction logic
│   └── schemas.py      # Pydantic request/response models
├── plots/              # Sample forecast visualizations
│   ├── AAPL_plot.png
│   ├── MSFT_plot.png
│   └── GOOG_plot.png
├── Dockerfile
├── .dockerignore
├── .gitignore
├── requirements.txt
├── LICENSE
└── README.md

Prophet vs. ARIMA vs. LSTM

Feature	Prophet	ARIMA	LSTM
Approach	Additive regression with Fourier seasonality	Statistical autoregressive model	Neural network with memory cells
Strengths	Handles missing data, holidays, changepoints automatically	Well-understood theory, works well on stationary data	Can learn complex nonlinear patterns
Weaknesses	Less control over model internals	Requires stationarity, manual parameter tuning	Needs large datasets, prone to overfitting
Best For	Business forecasting with strong seasonality	Short-term forecasting of stationary series	Large-scale sequential pattern learning

Prophet was chosen for this project because stock price data exhibits strong weekly and yearly seasonality, and Prophet handles this natively without manual seasonal decomposition.

License

MIT