280

Annexe 5

Figure 2. Four integration levels between Kate and Patdex

6. An integration architecture to handle missing values efficiently

As stated in section 3, one main drawback of a decision tree consultation occurs if the user answers "unknown" to a test. Unknown values propagate an uncertainty along all the branches of the "unknown node" - we define an unknown node as a node where the user answers "unknown" during the consultation of the tree although a subsequent test may remove this uncertainty. Moreover, the final diagnosis is probabilistic which is confusing for a non expert user. One way to deal with unknown values in the consultation of a tree is to switch to a case- based reasoning procedure after consulting the tree. When an unknown value is encountered, the consultation of the tree is stopped and the case-based reasoner is used to choose the next tests. The probabilistic diagnoses delivered by Kate may also be refined by using the similarity measure of the case-based reasoner. A workbench integration is needed. The procedure when encountering an unknown value in the consultation of the decision tree is presented below:
1.Get the current situation given by the first tests
of the tree.
2.Get the current subset of the cases listed under the
unknown node.
3.Switch to Patdex by using the current situation and
the current set of cases.

Procedure for Switching between Kate and Patdex

This procedure combines the advantages of both techniques for efficiency and correctness. In the worst case, the user answers unknown at the root node and we are left with a classical CBR consultation. In the best case, the user never answers unknown and we are left with a classical decision tree traversal mechanism that is very efficient.

Conclusions

Induction and case-based reasoning are complementary approaches for developing experience- based diagnostic systems. Induction compilespast experiences into general knowledge used to solve problems. Case-based reasoning directly interpretspast experiences. Both technologies