Title :
An empirical comparison of arc-cosine distance, generalised Fisher ratio and normalised entropy criteria for model selection
Author :
Zheng, S. ; Molina, C.G.
Author_Institution :
Dept. of Comput. Sci., Anglia Polytech. Univ., Cambridge, UK
fDate :
31 Aug-2 Sep 1998
Abstract :
Three model selection criteria, the arc-cosine distance, the generalised Fisher ratio and the normalised entropy, are applied to several data sets sampled from different mixture models. Their performance is investigated and their ability to measure the mutual information between the components in a mixture model is compared. Experimental results show that the arc-cosine distance criterion outperforms the other two criteria
Keywords :
entropy; modelling; pattern recognition; arc-cosine distance; data sets; generalised Fisher ratio; mixture models; model partitioning; model selection; normalised entropy criteria; Bayesian methods; Computer science; Entropy; Gaussian distribution; Monte Carlo methods; Mutual information; Partitioning algorithms; Performance evaluation;
Conference_Titel :
Neural Networks for Signal Processing VIII, 1998. Proceedings of the 1998 IEEE Signal Processing Society Workshop
Conference_Location :
Cambridge
Print_ISBN :
0-7803-5060-X
DOI :
10.1109/NNSP.1998.710651
