LINI · UAM Iztapalapa · Open-Access EEG Dataset
A large open-access EEG database from first-time users of a P300 Brain-Computer Interface speller, recorded at the Neuroimaging Research Laboratory of UAM Iztapalapa.
About the database
The Akimpech dataset was created at the Laboratorio de Investigación en Neuroimagenología (LINI) of the Universidad Autónoma Metropolitana, Unidad Iztapalapa, to support open research in Brain-Computer Interfaces based on the P300 event-related potential.
The P300 Speller, originally proposed by Farwell and Donchin (1988), is one of the most widely studied BCI paradigms. It relies on the P300 component — a positive EEG deflection ~300 ms after a rare, attended stimulus — to allow users to select characters from a matrix by focusing attention on them while rows and columns are intensified in random order.
The database was recorded using the BCI2000 P300 Speller platform and is available in both BCI2000 and MATLAB formats, along with support functions for classifier design and testing.
Protocol
Each of the 30 subjects completed 4 recording sessions. All sessions used 15 repetitions per character (except session 4 which could be reduced to 1).
Subjects spelled three fixed Spanish words: calor, cariño and sushi. Used to calibrate the BCI classifier. Training set: 16 characters with 15 epochs each.
Subjects spelled the word sushi. 15 repetitions per character.
Subjects spelled any word of their choice. 15 repetitions per character. Commonly used as the test set in classifier evaluations.
Subjects spelled freely, with the number of repetitions reduced to as few as 1 per character, exploring speed vs. accuracy trade-offs.
Most published studies using Akimpech train on Session 1 and test on Session 3. The dataset has been used with LDA, SVM, CNN, capsule networks and unsupervised learning approaches, among others.
Access the data
The full dataset is freely available through the following mirrors. Choose whichever works best for you.
Primary mirror. Full dataset including MATLAB files and support functions.
Download ↗Original LINI server at UAM Iztapalapa with dataset documentation.
Browse dataset ↓ Download ZIP ⬇Community mirror on Kaggle. May require a free Kaggle account.
View on Kaggle ↗If you cannot access the data through any of the mirrors above, write directly to oyanez@izt.uam.mx and we will arrange access.
Reference
If you use the Akimpech dataset in your research, please cite the original conference paper:
Plain text: Ledesma-Ramirez C., Bojorges-Valdez E., Yáñez-Suarez O., Saavedra C., Bougrain L., Gentiletti G.G. (2010). "An Open-Access P300 Speller Database." Fourth International Brain-Computer Interface Meeting, Monterey, CA, USA.
Research team
For enquiries about the dataset contact oyanez@izt.uam.mx or erik.bojorges@ibero.mx.
Individual files
Browse and download individual files for each subject. Click a subject row to expand its sessions. Each session lists all available files — click any chip to download directly.
Impact
The following is a selection of peer-reviewed publications that have used the Akimpech dataset for algorithm development, benchmarking, or analysis. If your work uses this dataset and is not listed, please contact us.
Toward Lightweight, Efficient, and Explainable Deep Learning Models for P300 Detection
Merging Brain-Computer Interface P300 Speller Datasets: Perspectives and Pitfalls
Improving P300 Speller Performance by Means of Optimization and Machine Learning
Enhancing P300-Based Character Recognition Performance Using a Combination of Ensemble Classifiers and a Fuzzy Fusion Method
Recurrence Analysis of the P300 Response During a BCI Spelling Task
Wavelet-Based Semblance Methods to Enhance the Single-Trial Detection of Event-Related Potentials for a BCI Spelling System
A Comparison of Subject-Dependent and Subject-Independent Channel Selection Strategies for Single-Trial P300 Brain-Computer Interfaces
P300 Detection Based on EEG Shape Features
A Bayesian Model for Exploiting Application Constraints to Enable Unsupervised Training of a P300-based BCI
Finally, What Is the Best Filter for P300 Detection?
A P300 BCI for the Masses: Prior Information Enables Automatic Language Model Based Unsupervised Adaptation
Used the Akimpech dataset in your research? Write to oyanez@izt.uam.mx or erik.bojorges@ibero.mx to have your publication listed here.