skforecast is a Python library for time series forecasting built on the scikit-learn API. It turns any scikit-learn compatible regressor (LightGBM, XGBoost, CatBoost and many more), statistical models (ARIMA, SARIMAX, ETS) and pre-trained foundation models (Chronos-2, TimesFM, Moirai-2 and others) into forecasters that share one API, for one series or thousands of related series. Typical uses are forecasting energy demand, sales, web traffic or intermittent demand.
A reliable forecast needs much more than a model: lag and window features, exogenous variables, backtesting that does not leak future data into the past, hyperparameter search, prediction intervals and drift detection in production. skforecast provides them as tested and documented building blocks, so data scientists in industry and research do not have to reimplement them, and their subtle errors, in every project.
What sets it apart: it extends the scikit-learn ecosystem instead of replacing it, so teams keep the models, transformers and tools they already use, and it evaluates machine learning, statistical and foundation models with the same backtesting, so they can be compared fairly on the same data. Its free documentation and tutorials, in English, Spanish and Chinese, bring forecasting to people who prefer to learn in their own language.
skforecast is downloaded about 95,000 times a month from PyPI, around 4 million times in total, and more than 70 scientific publications use it. A forecasting pipeline built on it cannot be moved to another library without rewriting it, so its users depend on it staying compatible with every new release of its dependencies. That continuous maintenance is what this funding pays for.
A reliable forecast needs much more than a model: lag and window features, exogenous variables, backtesting that does not leak future data into the past, hyperparameter search, prediction intervals and drift detection in production. skforecast provides them as tested and documented building blocks, so data scientists in industry and research do not have to reimplement them, and their subtle errors, in every project.
What sets it apart: it extends the scikit-learn ecosystem instead of replacing it, so teams keep the models, transformers and tools they already use, and it evaluates machine learning, statistical and foundation models with the same backtesting, so they can be compared fairly on the same data. Its free documentation and tutorials, in English, Spanish and Chinese, bring forecasting to people who prefer to learn in their own language.
skforecast is downloaded about 95,000 times a month from PyPI, around 4 million times in total, and more than 70 scientific publications use it. A forecasting pipeline built on it cannot be moved to another library without rewriting it, so its users depend on it staying compatible with every new release of its dependencies. That continuous maintenance is what this funding pays for.