Abstract
Aim:
To propose a modification of the TWINSPANalgorithm that enables production of divisive classifi-cations that better respect the structure of the data.
Methods:
The proposed modification combines the classi-cal TWINSPAN algorithm with analysis of heterogeneityof the clusters prior to each division. Four differentheterogeneity measures are involved: Whittaker’s beta,total inertia, average S
rensen dissimilarity and averageJaccard dissimilarity. Their performance was evaluatedusing empirical vegetation datasets with different numbersof plots and different levels of heterogeneity.
Results:
While the classical TWINSPAN algorithm divideseach cluster coming from the previous division step, themodified algorithm divides only the most heterogeneouscluster in each step. The four tested heterogeneity measuresmay produce identical or very similar results. However,average Jaccard and S
rensen dissimilarities may reachextreme values in clusters of small size and may produceclassifications with a highly unbalanced cluster size.
Conclusions:
The proposed modification does not alter thelogic of the TWINSPAN classification, but it may changethe hierarchy of divisions in the final classification. Thus,unsubstantiated divisions of homogeneous clusters areprevented, and classifications with any number of terminalclusters can be created, which increases the flexibility of TWINSPAN
Abstract
Aim:
To propose a modification of the TWINSPANalgorithm that enables production of divisive classifi-cations that better respect the structure of the data.
Methods:
The proposed modification combines the classi-cal TWINSPAN algorithm with analysis of heterogeneityof the clusters prior to each division. Four differentheterogeneity measures are involved: Whittaker’s beta,total inertia, average S
rensen dissimilarity and averageJaccard dissimilarity. Their performance was evaluatedusing empirical vegetation datasets with different numbersof plots and different levels of heterogeneity.
Results:
While the classical TWINSPAN algorithm divideseach cluster coming from the previous division step, themodified algorithm divides only the most heterogeneouscluster in each step. The four tested heterogeneity measuresmay produce identical or very similar results. However,average Jaccard and S
rensen dissimilarities may reachextreme values in clusters of small size and may produceclassifications with a highly unbalanced cluster size.
Conclusions:
The proposed modification does not alter thelogic of the TWINSPAN classification, but it may changethe hierarchy of divisions in the final classification. Thus,unsubstantiated divisions of homogeneous clusters areprevented, and classifications with any number of terminalclusters can be created, which increases the flexibility of TWINSPAN
การแปล กรุณารอสักครู่..