Induction of Uncertain Rules and the Sociopathicity Property in Dempster-Shafer Theory

As is well known, Dempster-Shafer theory offers an alternative approach to deal with uncertainty reasoning in expert systems and other fields. In this paper, we present two results that relate to the use of the theory. First, we present and analyze four methods to induce uncertain rules from a training instance set for Dempster-Shafer theory. This is the first attempt to do so, to the best of our knowledge. Second, we show that these “correct” induced rules exhibit the sociopathicity property when they are used in evidential reasoning. The sociopathicity property states that a system's performance can degrade when some individually good pieces of uncertain knowledge (in terms of rules) are added to the knowledge base. The importance of this result is that the performance of evidential reasoning systems based on Dempster-Shafer theory can deteriorate for essentially any size of knowledge base, although all the pieces of knowledge in the knowledge base are “correct”.