JavaQCredo: A Software for Qualitative Credal Networks

Credal networks enhance robustness and modelling power of Bayesian networks by allowing for the specication of credal sets (i.e., convex sets of probability mass functions), instead of single conditional mass functions in the model denition [2]. Unlike the case of Bayesian networks, for which a large number of software tools with graphical interface are available (e.g., Hugin, Netica, Samiam, JavaBayes), no tools of this kind have been developed so far for credal networks. The Java software we present, called JavaQCredo, and based on Fabio Cozman’s code originally developed for JavaBayes, should be regarded as a rst step in this direction. Its implementation has been granted by the Swiss army, with the goal of supporting military experts in their strategic decision tasks. Model variables and conditional dependencies are associated to nodes and arcs of a graph in the same way as any other tool for Bayesian networks. The distinctive feature of JavaQCredo concerns the quantication of the conditional models for the dierent variables: no numbers have to be explicitly assessed to express expert knowledge [3]. The user is simply required to formulate qualitative judgements about the (conditional) states of the variables like, for instance, \if Y is y, X is likely to be x". This is achieved by scrolling drop-down lists to select the quantier