Comparing three ML approaches to time-series stock price forecasting across 5 major tech tickers: AAPL, MSFT, GOOGL, TSLA, AMZN.
| Ticker | Prophet RMSE | XGBoost RMSE | LSTM RMSE | Best Model |
|---|---|---|---|---|
| AAPL | 25.20 | 22.04 | 14.48 | LSTM |
| MSFT | 93.27 | 39.17 | 16.24 | LSTM |
| GOOGL | 52.67 | 97.57 | 16.00 | LSTM |
| TSLA | 48.12 | 21.43 | 19.75 | LSTM |
| AMZN | 24.39 | 15.33 | 11.27 | LSTM |
- XGBoost achieved the strongest directional accuracy on 3 of 5 tickers
- LSTM captured longer-term trend reversals better than XGBoost
- Prophet was most interpretable — clear trend/seasonality decomposition
- TSLA showed the lowest directional accuracy across all models (highest volatility — consistent with its beta)
This project is for educational and portfolio purposes only. It is not financial advice and is not intended for live trading.
| File | Description |
|---|---|
| notebooks/01_eda.ipynb | EDA + technical indicators |
| notebooks/02_prophet.ipynb | Prophet trend forecasting |
| notebooks/03_xgboost.ipynb | XGBoost with engineered features |
| notebooks/04_lstm.ipynb | LSTM neural network |
| src/data_loader.py | yfinance data pipeline |
| src/features.py | Technical indicator engineering |
| src/evaluate.py | Shared evaluation metrics |
Python | yfinance | TensorFlow/Keras | Prophet | XGBoost | pandas | Plotly
git clone https://github.com/Kay-Lander/stock-forecasting.git cd stock-forecasting python -m venv venv && venv\Scripts\activate pip install -r requirements.txt python src/data_loader.py # fetch and cache stock data jupyter notebook # run notebooks in order 01 → 04