aeon-neuro

Reproducible EEG classification, built on modern time series machine learning.

Documentation · Source code · PyPI · Discussions

EEG classification research is fragmented. Methods are spread across signal processing, deep learning, BCI, and general machine learning toolkits, often with different data formats, bespoke pipelines, and evaluation protocols.

aeon-neuro brings these strands together in a single open-source package for EEG time series classification. It is designed to make experiments easier to run, easier to compare, and easier to reproduce. The project provides a unified interface for EEG classification problems, benchmark datasets, and baseline methods, while building on the design principles of aeon, the Python toolkit for time series machine learning.

Whether you are testing a new EEG method, benchmarking against strong baselines, or trying to reproduce published results, aeon-neuro aims to give you a cleaner starting point.

Overview

Category	Badges
CI/CD
Code
Community

Why aeon-neuro?

A unified interface for EEG classification

Work with EEG classification problems through a consistent API instead of stitching together multiple incompatible workflows.

Reproducible benchmarks

Run experiments on curated EEG classification problems with fixed train/test splits and shared evaluation protocols.

Strong baselines out of the box

Compare against established methods from time series machine learning, deep learning, and related EEG workflows.

Built with open science in mind

The package is developed to support transparent, reproducible, and extensible research.

Part of the aeon ecosystem

aeon-neuro is a companion package to aeon, extending its time series machine learning foundations to EEG applications.

What this project is for

aeon-neuro is intended for researchers and practitioners who want to:

• benchmark new EEG classifiers against strong baselines,
• reproduce published EEG classification experiments,
• load and evaluate EEG benchmark problems consistently,
• test general-purpose time series classifiers on EEG data,
• build more reliable experimental pipelines for EEG research.

The project is especially motivated by a simple problem in the field: EEG classification is widely studied, but comparisons are often hard because data preparation, feature engineering, model choices, and evaluation protocols vary so much from paper to paper.

Benchmark-first design

A central goal of aeon-neuro is to support fairer and easier evaluation.

The project is tied to a benchmark archive of EEG classification problems spanning medical, brain-computer interface, and psychology applications. This makes it possible to compare methods across a broader range of tasks, rather than drawing conclusions from a small hand-picked subset of datasets.

This also makes aeon-neuro useful beyond software alone. It is a research scaffold for building stronger baselines, reproducing studies, and understanding where different classes of method work well or fail.

Installation

aeon-neuro requires Python 3.9 or later. Full installation instructions are available in the documentation.

Install the core package with pip:

pip install aeon-neuro

To install with all optional dependencies:

pip install aeon-neuro[all_extras]

To install the latest development version from GitHub, see the installation guide.

Documentation

• Project documentation: https://aeon-neuro.readthedocs.io
• aeon documentation: https://aeon-toolkit.org
• Source code: https://github.com/aeon-toolkit/aeon-neuro

Where to ask questions

Contributing

We welcome contributions across code, datasets, benchmarking, documentation, and testing.

If you want to contribute, start with:

• the issue tracker,
• the discussions page,
• or the community Slack.

Acknowledgements

This work is supported by the UK Engineering and Physical Sciences Research Council (EPSRC) under grant EP/W030756/2.