Historical features
Lagged closes, rolling averages and volatility, plus 5-, 10-, and 20-session momentum features.
Open-source ML project
Explore prices.
Explore prices.
Test predictions.
A Python machine-learning project that compares Linear Regression and Random Forest on historical stock prices, with expanding walk-forward evaluation instead of random train/test splits.
Built with Python · scikit-learn · yfinance · Streamlit
Model evaluation
Compare before you trustMAEAverage absolute error
RMSEPenalizes large errors
R²Relative fit score
Illustrative graphic only — no live market data or forecast is shown on this static page.
Built to learn, measure, and compare.
The goal is not to promise a winning trade. It's to make the modeling process inspectable and compare predictions against a simple baseline.
Two models + baseline
Compare Linear Regression and Random Forest against a previous-close baseline on later observations.
Time-aware evaluation
Three expanding chronological windows with MAE, RMSE, R², directional accuracy, and an actual-vs-predicted chart.
How the model works
A clear, repeatable pipeline from price history to a one-session forecast.
STEP 01
Get price history
Download adjusted daily closing prices through yfinance.
STEP 02
Build features
Construct lagged and rolling features using prior observations.
STEP 03
Test honestly
Expand the training history and evaluate on later dates, without shuffling.
STEP 04
Compare errors
Measure model errors and compare them with a naive baseline.
Use the predictor online
Deploy the existing Python app to Streamlit Community Cloud. After the one-time deployment, the app runs in the cloud and visitors can use it from a browser without installing Python locally.
1. Deploy to Streamlit
Sign in with GitHub, review the repository settings, and deploy app.py from the main branch.
First deployment requires your own Streamlit/GitHub authorization. Choose a public app if you want others to use it.
2. What the online app does
Users enter a Yahoo Finance ticker, choose 1, 2, 5, or 10 years of daily history, and run an online forecast. The app compares Linear Regression and Random Forest with a previous-close baseline across expanding chronological test windows.
Data timing: The app uses completed daily adjusted closes, omits the current session until its regular close, and caches data for up to one hour. Yahoo Finance availability and delays may vary; this is not a real-time exchange feed. Forecasts are estimates, not buy/sell signals.
Want to run it locally instead?
Install dependencies
python -m venv .venv python -m pip install -r requirements.txt
Start Streamlit
python -m streamlit run app.py