276

Annexe 5

function is adapted using a connectionist learning technique (competitive learning). For the test selection, the adaptation of similarity measures is based on an estimation of the average costs for ascertaining symptoms using an A*-like procedure. PATDEX can deal with redundant, incomplete, and incorrect cases and includes the processing of uncertain knowledge through default values. PATDEX is described in [4] and [5].

3

The need for integration

INRECA integrates induction and case-based reasoning so that they can collaborate and provide better solutions than they would individually. Before describing how integration is performed, we first state why the two approaches are complementary. Induction presents some limitations for building an identification system that can handle missing values during consultation. Consider the following case base drawn from an application that identifies marine sponges developed at the Museum of Natural History in Paris.

Table 1 - A database of cases for an application which identifies marine sponges

KATE works in two steps: it first learns a decision tree and then uses the tree to identify the unknown class of a new incoming sponge. Consider what happens when the user does not know how to answer the first question asked during consultation of the tree of figure 1.

When the user answers "unknown", KATE proceeds by following both branches "lancet- shape" and "large" and combines the conclusions found at the leaves. In the "large" branch, it reaches the "Paradisconema" leaf node. In the "lancet-shape" branch, it reaches a test node and "body". He answers "conical".

Figure 1:A consultation of the decision tree learned by KATE

the user is queried for the value of the "shape" of the object

KATE

reaches the "Coscinonema" leaf and combines the two

leaves to conclude that the current case is a "Paradisconema" with a probability of 0.5 or a "Coscinonema" with a probability of 0.5. Consider case ex1 at the "Paradisconema" leaf node. The feature "shape(body)" of ex1 has the value "ellipsoid" unlike the current case where it is "conical". Thus, the current case is closer to ex2 than to ex1 and the correct conclusion is "Coscinonema" with a probability of 1. Unfortunately, the information about the "body shape" of ex1 was generalized away during induction and is no longer available during consultation.

Note that there are other methods for handling unknown values during consultation of a tree. Instead of combining branches, one can assign a probability to the branches [6] and follow the