Classification using distance-based segmentation—application to the analysis of EEG signals

The algorithm described in this paper first analyses the trajectory of observations in the space to distinguish observations corresponding to classes from observations corresponding to transition sequences between classes. This first step is called ‘segmentation’ since it arranges data into segments: class segments and transition segments. Segments are then merged using a hierarchical clustering method until classes and transition sequences are interlaced. This algorithm has been applied to classify sleep EEG samples.