Python Versions PyPI Documentation Status Binder DOI

⭐️ Simple. Robust. Powerful.

FLIRT is a feature generation toolkit for wearable data such as that from your smartwatch or smart ring. With FLIRT you can easily transform wearable data into meaningful features which can then be used for example in machine learning or AI models.

In contrast to other existing toolkits, FLIRT (1) focuses on physiological data recorded with (consumer) wearables and (2) calculates features based on a sliding-window approach. FLIRT is an easy-to-use, robust and efficient feature generation toolkit for your wearable device!

FLIRT Workflow

➡️ Are you ready to FLIRT with your wearable data?

Main Features

A few things that FLIRT can do:

  • Loading data from common wearable device formats such as from the Empatica E4 or Holter ECGs

  • Overlapping sliding-window approach for feature calculation

  • Calculating HRV (heart-rate variability) features from NN intervals (aka inter-beat intervals)

  • Deriving features for EDA (electrodermal activity)

  • Computing features for ACC (accelerometer)

  • Provide and prepare all features in one comprehensive DataFrame, so that they can directly be used for further steps (e.g. training machine learning models)

😎 FLIRT provides high-level implementations for fast and easy utilization of feature generators (see flirt.simple).

🤓 For advanced users, who wish to adapt algorithms and parameters to their needs, FLIRT also provides low-level implementations. They allow for extensive configuration possibilities in feature generation and the specification of which algorithms to use for generating features.


FLIRT is available from PyPI and can be installed via pip:

pip install flirt

See Installation for further info.

Quick example

Generate a comprehensive set of features for an Empatica E4 data archive with a single line of code 🚀

import flirt
features = flirt.with_.empatica('./')

Check out the exemplary Jupyter notebooks.


Made with ❤️ at ETH Zurich.

Check out all Authors.


  • How does FLIRT distinguish from other physiological data processing packages such as neurokit? While FLIRT works with physiological data like other packages, it places special emphasis on the inherent challenges of data processing obtained from (consumer) wearable devices such as smartwaches instead of professional, medical-grade recording devices such as ECGs or EEGs. As an example, when processing data from smartwatches, one could be confronted with inaccurate data, which needs artifact removal, or measurement gaps, which need to be dealt with.


Original article: FLIRT: A Feature Generation Toolkit for Wearable Data

    title={{{FLIRT}}: A {{Feature Generation Toolkit}} for {{Wearable Data}}},
    author={Föll, Simon and Maritsch, Martin and Spinola, Federica and Mishra, Varun and Barata, Filipe and Kowatsch, Tobias and Fleisch, Elgar and Wortmann, Felix},
    journal={Computer Methods and Programs in Biomedicine},

Indices and tables