an:01163858
Zbl 0902.62006
Ciampi, A.; Diday, E.; Lebbe, J.; P??rinel, E.; Vignes, R.
Tree-growing with probabilistically imprecise data
EN
Diday, Edwin (ed.) et al., Ordinal and symbolic data analysis. Proceedings of the international conference, OSDA 95, Paris, June 20--23, 1995. Berlin: Springer. Studies in Classification, Data Analysis, and Knowledge Organization. 201-212 (1996).
1996
a
62-07
binary segmentation; decision tree; algorithm of tree-growing; imprecise data; symbolic data analysis; recursive partition; maximum likelihood criterion; mixture distributions problem; EM algorithm
Summary: An algorithm of tree-growing is proposed to treat explicitely imprecise data such as error measurements or subjective probabilistic judgments. These are formally described with probabilistic assertions in the framework of symbolic data analysis. In this context, the recursive partition procedure can be viewed as an iterative search for an organized set of symbolic objects which best fits the initial data. At each step, the best split is obtained under a maximum likelihood criterion. The imprecision induces a mixture distributions problem which is solved through the EM algorithm. A medical example illustrates this approach.
For the entire collection see [Zbl 0866.00054].