CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Sparse PLS-DA: Clustering time series for art conservation Authors:  Sandra Ramirez - Universidad Politecnica de Valencia (Spain) [presenting]
Manuel Zarzo - Universidad Politecnica de Valencia (Spain)
Fernando-Juan Garcia-Diego - Universidad Politecnica de Valencia (Spain)
Angel Perles - Universidad Politecnica de Valencia (Spain)
Abstract: Clustering time series data has a wide range of applications. One problem when analyzing time series for art conservation is that time series of relative humidity RH are too similar, even when positioned differently. Many studies have displayed that one common problem is that if underlying clusters are very close to each other, the clustering performance might diminish significantly. Before applying the discriminant technique, the variables that are extracted from the time series were determined. They correspond to estimates of parameters from time series models and features from time series functions. The number of variables was greater than the number of time series. The goal is to propose a methodology for classifying time series when they are similar and have more variables. Sparse Partial Least Squares Discriminant Analysis (sPLS-DA) was applied using three prediction distances and two classification error rates. The methodology is described using different time series of RH from a study carried out in the Metropolitan Cathedral of Valencia in 2008 and 2010. This methodology would be perfect for applying to time series data of relative humidity from both churches and museums, or similar buildings where it is possible to indicate the class of the time series.