Data Mining Portfolio

Introduction

This dataset was first made available at the Brain-Computer Interface Competition III as part of an effort to validate signal processing and classification methods for Brain-Computer Interfaces.

Provided by José del R. Millán from the IDIAP Research Institude, his challenge to the participants was to give a description of the algorithm applied to the dataset to determine the class labels based on a given training and testing dataset.

Goal of Data Mining

I will trying to assign class labels to the test data to indicate which activity the subject was performing while the data were collected. This goal should be attained by training a minimum of two different classifiers.

Data Collection

EEG data was collected from three participants while each imagined moving their left hand, moving their right hand, and thinking of words that start with the same letter. Each subject performed the given task for approximately 15 seconds and then changed tasks. For each subject, the raw EEG measurements and a precomputed set of values are proved.

Each line in the raw measurements file contains the EEG potentials from the following probes in this order: Fp1, AF3, F7, F3, FC1, FC5, T7, C3, CP1, CP5, P7, P3, Pz, PO3, O1, Oz, O2, PO4, P4, P8, CP6, CP2, C4, T8, FC6, FC2, F4, F8, AF4, Fp2, Fz, Cz. A class label is provided with the training data.

The precomputed values contain the raw EEG measurements that were filtered spacially via a surface Laplacian function. The power spectral density (PSD) in the 8-30 Hz bandwidth was estimated 16 times per second for eight centro-parietal channels: C3, Cz, C4, CP1, CP2, P3, Pz, and P4. Again, a class label was provided with the training data.

Next: Tools
Back to: Main Page