From 3a5a5a09346ac4b66e744ce94907bc1ce663ca6a Mon Sep 17 00:00:00 2001 From: unintendedbear Date: Wed, 30 Aug 2017 15:16:31 +0200 Subject: [PATCH] Reference missing. References #11 --- Chapters/03-softc.tex | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/Chapters/03-softc.tex b/Chapters/03-softc.tex index ac2d555..c6821d7 100644 --- a/Chapters/03-softc.tex +++ b/Chapters/03-softc.tex @@ -118,4 +118,4 @@ \subsection{Data visualisation and interpretation} The process of forming the clusters is by measuring how alike the observations are. The algorithm k-means \cite{hartigan1975clustering} has been traditionally used for creating \textit{k} clusters, and filling them depending on the Euclidean distances between the center of the clusters and the rest of the data. Next, the \textit{mean} of the instances is calculated and set as the new center of every cluster, and the rest of the data ir re-assigned again. The process finishes when there are no changes in the clusters. -Another tool that can be used for interpretation, as well as for classification, is \textit{association learning}. Association learning is also useful when the classes have not been specified. This kind of algorithms can predict not only the class but any attribute, which is why they are also used to extract interesting knowledge more than for only predicting. The same way the goodness of a classifier model is expressed by their accuracy over a test set of data, the association rules are accompanied by their \textit{confidence}, meaning te percentage of the observations over the total that the rule represents correctly. And therefore, by discovering the associations among the instances, one can visualise the relationships between the observations, similarly to what the clusters do. \ No newline at end of file +Another tool that can be used for interpretation, as well as for classification, is \textit{association learning}. Association learning is also useful when the classes have not been specified. This kind of algorithms \cite{hipp2000algorithms} can predict not only the class but any attribute, which is why they are also used to extract interesting knowledge more than for only predicting. The same way the goodness of a classifier model is expressed by their accuracy over a test set of data, the association rules are accompanied by their \textit{confidence}, meaning te percentage of the observations over the total that the rule represents correctly. And therefore, by discovering the associations among the instances, one can visualise the relationships between the observations, similarly to what the clusters do. \ No newline at end of file