ABSTRACT
The general goal of clustering is to identify structure in an unlabeled data set by organizing the data into homogeneous groups with maximum similarity within a group and the largest dissimilarity between groups. There has been growing interest in using time series clustering as a data mining tool in natural, life, and social sciences. In this chapter, we present a methodology that can choose a set of relevant spectral (spectrum-related) features for clustering segments of time series with similar characteristics in an automatic manner and determine the optimal number of clusters at the same time. In our application, we use electroencephalogram (EEG) signals collected from a patient with epilepsy. Our method is able to distinguish the epileptic seizure EEG signals from the normal state EEG signals.
