Development of Possibilistic Causal Model from Data

Possibilistic causal models have been proposed as an approach for prediction and diagnosis based on uncertain causal relations. However, the only way to develop the causal models is to acquire the possibilistic knowledge from the experts. The paper proposes an approach to develop the models from a dataset including causes and effects. It first develops a probabilistic causal model, then transform it into a possibilistic one. The points which should be discussed in the approach are 1) the way to transform multiple probabilistic distributions consistently into possibilistic ones, and 2) the merits of the transformation from a probabiistic model into a possibilistic one. Key-Words: Causal model, possibility theory, probability-possibility transformation, machine learning.