Imputation of Possibilistic Data for Structural Learning of Directed Acyclic Graphs
One recent focus of research in graphical models is how to learn them from imperfect data. Most of existing works address the case of missing data. In this paper, we are interested by a more general form of imperfection i.e. related to possibilistic datasets where some attributes are characterized by possibility distributions. We propose a structural learning method of Directed Acyclic Graphs (DAGs), which form the qualitative component of several graphical models, from possibilistic datasets. Experimental results show the efficiency of the proposed method even in the particular case of missing data regarding the state of the art Closure under tuple intersection (CUTS) method .
KeywordsBayesian Network Directed Acyclic Graph Maximum Frequency Possibility Distribution Possibility Theory
Unable to display preview. Download preview PDF.
- 1.Borgelt, C., Steinbrecher, M., Kruse, R.: Graphical models: representations for learning, reasoning and data mining, vol. 704. Wiley (2009)Google Scholar
- 2.Pearl, J.: Probabilistic reasoning in intelligent systems: networks of plausible inference. Morgan Kaufmann (1988)Google Scholar
- 8.Fonck, P.: Propagating uncertainty in a directed acyclic graph. In: IPMU, vol. 92, pp. 17–20 (1992)Google Scholar
- 9.Ayachi, R., Ben Amor, N., Benferhat, S., Haenni, R.: Compiling possibilistic networks: Alternative approaches to possibilistic inference. In: Proceedings of the Twenty-Sixth Conference Annual Conference on Uncertainty in Artificial Intelligence (UAI 2010), Corvallis, Oregon, pp. 40–47. AUAI Press (2010)Google Scholar