The AI framework
for Time Series

  • Forecast
  • Detect
  • Classify
  • open source
  • no lock-in
  • cutting edge
  • user led
  • sovereign
  • business ready
forecast_output.pyexamples

Hourly energy demand

Forecast load to balance the grid and cut peak costs.

HistoricalForecast
For enterprise

Plan better, spend less

We implement and run sktime inside your company.

  • Production-grade AI
  • SLA & support
  • Forward-deployed engineers
  • Training & consulting
Explore enterprise
For developers

Get started in minutes

One install, one API, every model.

  • Unified API
  • 500+ models
  • Open source
  • Permissive license
Read the quickstart
One framework, every model

Many AI models.
One harness.

Classical statistics, deep learning and modern AI foundation models. sktime gives you every approach behind a single, consistent API. Pick the right model for each task, and switch anytime without rewriting your code.

  • 500+ models, one API

    Classical statistics, machine learning, deep learning, and modern AI, all behind the same interface.

  • Switch models in one line

    Swap algorithms to find what works best, without rewriting your pipeline or learning a new tool.

  • Modern AI, no lock-in

    Run AI foundation models next to proven classics, same API, no lock-in. Open source, so your data stays yours.

  • Add private components easily

    Bring your own estimators, transformers, or adapters from a closed codebase and use them through the same sktime API.

forecast.py
from sktime.forecasting.arima import AutoARIMA
from sktime.datasets import load_airline

y = load_airline()  # your time series
forecaster = AutoARIMA(sp=12)
forecaster.fit(y)  # train on history
y_pred = forecaster.predict(fh=[1, 2, 3])

Swapping a model is a two-line edit. Your load, fit, and predict stay the same.

Browse all 500+ models
Capabilities

One framework,
every time series task

A unified Python API for the complete time series machine learning lifecycle, from exploration to production.

Forecasting

Predict future demand, revenue, and capacity so teams can plan ahead with confidence.

Feature Extraction

Turn raw time series into clean, model-ready inputs without slow manual preparation.

Probabilistic Prediction

Get forecasts with confidence ranges, not just single numbers, so you can weigh risk and uncertainty.

Anomaly and Change Detection

Catch equipment failures, fraud, and sudden shifts the moment they appear in your data.

Classification

Sort time series into categories automatically. Signal types, product groups, or machine states.

Time-to-Event Modeling

Estimate when an event will happen — churn, failure, or conversion — and how likely it is.

Regression

Estimate a number from a sequence, such as remaining equipment life or expected output.

Clustering

Group similar time series to reveal customer segments, product families, and hidden patterns.

Alignment

Compare and match sequences that run at different speeds or start at different times.

For developers · Open source

Start building in minutes

Install sktime and go from raw data to a forecast with a familiar, scikit-learn-style API.

Install
$ pip install sktime

New to time series?

No problem. The getting-started guide explains the core ideas and walks you through your first project step by step.

Start with the basics
See it in action

Five lines from data
to prediction

sktime's unified API makes time series forecasting as simple as scikit-learn makes classification. Load data, fit a model, predict the future.

  • Consistent API across 500+ models
  • Swap models without changing your code
  • Built on scikit-learn design patterns
forecast.py
 1  from sktime.forecasting.chronos import ChronosForecaster
 2  from sktime.datasets import load_airline
 3
 4  y = load_airline()              # monthly passenger data
 5  forecaster = ChronosForecaster(
 6      "amazon/chronos-bolt-tiny"
 7  )                            # create forecaster
 8  forecaster.fit(y)               # learn from history
 9  y_pred = forecaster.predict(   # predict next 3 months
10      fh=[1, 2, 3]
11  )

Built by the community, for the community

sktime is free and open source under the BSD-3-Clause license. Join hundreds of contributors shaping the future of time series in Python.

Enterprise Solutions

Need time series AI tailored to your business?

Custom features, infrastructure integration, and dedicated support - so you can focus on decisions, not pipelines.