Time Series Clustering from High Dimensional Data

Due to technological advances there is the possibility to collect datasets of growing size and dimension. On the other hand, standard techniques do not allow the easy management of large dimensional data and new techniques need to be considered in order to find useful results. Another relevant problem is the information loss due to the aggregation in large data sets. We need to take into account this information richness present in the data which could be hidden in the data visualization process. Our proposal - which contributes to the literature on temporal data mining - is to use some new types of time series defined as the beanplot time series in order to avoid the aggregation and to cluster original high dimensional time series effectively. In particular we consider the case of high dimensional time series and a clustering approach based on the statistical features of the beanplot time series.