| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 133.49 KB | Adobe PDF |
Advisor(s)
Abstract(s)
In this paper we present a model-based approach to the clustering of interval data building on recently
proposed parametric models. These methods consider configurations for the variance-covariance matrix that
take the nature of the interval data directly into account. The proposed framework relies on parametrizations
considering the inherent variability of the relevant data units and the relation that may exist between
this variability and the corresponding value levels. Using both synthetic and real data sets the pertinence of
the proposed methodology is shown, as the method effectively selects heterocedastic models with restricted
covariance structures when they are the most suitable, even in situations with limited information. Moreover,
considering special configurations of the variance-covariance matrix, adapted to nature of interval data,
proves to be the adequate approach. The presented study also makes clear the need to consider both the information about position (conveyed by the MidPoints) and intrinsic variability (conveyed by the Log-Ranges)
when analysing interval data.
Description
Keywords
Cluster Iinite mixture models interval-valued variable Intrinsic variability Symbolic data
Pedagogical Context
Citation
BRITO, Paula; DUARTE SILVA, A.P.; DIAS, José G.- Identifying Special Structures in Interval-Data via Model-Base Clustering. In 59th ISI World Statistics Congress, Hong Kong, China, 20-30 August, 2013. - In Proceedings 59th ISI World Statistics Congress, Netherlands: International Statistical Institute, 2013. ISBN 978-90-73592-34-6. p. 3899-3904