A simulation study : hierarchical PLS for multi-group classification
Master thesis
View/ Open
Date
2015-08-05Metadata
Show full item recordCollections
- Master's theses (KBM) [888]
Abstract
Hierarchically Ordered Taxonomic Partial Least Squares (Hot PLS) is a method for classifying data in a hierarchical structure. Since Hot PLS is a relatively new method, we want to study strengths and weaknesses of this. This was done by simulated data with known parameters by using the R package, Simrel.
The simulated data was then classified by Hot PLS. Classification error was used as the measure on how good the a method is to classify the data. For finding out which effect the different simulated parameters had on the classification error an ANOVA model was made, where the classification error was the response and the simulatated parameters and methods was the treatments. The simulated data were also classifies by other classifiers PLS, LDA, QDA and KNN, so one could check if the Hot PLS did perform better than the other classifiers. First the Hot PLS was only compared with PLS.
The results from these analysis show us that the Hot PLS is a good method for classifying data which has a hierarchical structure.