Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

techniques, we first introduce the notion of imprecision and uncertainty and
their possible sources with examples. Consider a problem of medical
diagnosis. Here, the pieces of knowledge in the knowledge base describe a
mapping from symptom space to disease space. For example, one such piece
of knowledge, represented by production rules, could be
Rule: IF Has-fever (Patient) AND
Has-rash (Patient) AND
Has-high-body-ache (Patient)
THEN Bears-Typhoid (Patient).
Now, knowing that the patient has fever, rash and high body pain, if the
diagnostic system infers that the patient is suffering from typhoid, then the
diagnosis may be the correct one (if not less) in every hundred cases. A
question then naturally arises: is the knowledge base incomplete? If so, why
don’t we make it complete? In fact, the above piece of knowledge suffers from
the two most common forms of incompleteness: firstly, there is a scope of
many diseases with the same symptoms, and secondly, the degree or level of
the symptoms is absent from the knowledge. To overcome the first problem,
the knowledge engineer should design the knowledge base with more specific
rules (i.e., rules with maximum number of antecedent clauses, as far as
practicable; see the specificity property in chapter 3). For rules with identical
symptoms (antecedent clauses), some sort of measures of coupling between the
antecedent and the consequent clauses are to be devised. This measure may
represent the likelihood of the disease, as depicted in the rule among its
competitive disease space. Selection of the criteria for this coupling may
include many issues. For example, if the diseases are seasonal, then the
disease associated with the most appropriate season may be given a higher
weightage, which, in some ways should be reflected in the measure of
coupling. The second problem, however, is more complex because the setting
of the threshold level at the symptoms to represent their strength is difficult
even for expert doctors. In fact, the doctors generally diagnose a disease from
the relative strength of the symptoms but quantification of the relative levels
remains a far cry to date. Besides the above two problems of incompleteness /
inexactness of knowledge, the database too suffers from the following kinds of
problems. Firstly, due to inappropriate reporting of facts (data) the inferences
may be erroneous. The inappropriate reporting includes omission of facts,
inclusion of non-happened fictitious data, and co-existence of inconsistent
data, collected from multiple sources. Secondly, the level or strength of the
facts submitted may not conform to their actual strength of happening, either
due to media noise of the communicating sources of data or incapability of the
sources to judge the correct level/ strength of the facts. Further, the observed
data in many circumstances do not tally with the antecedent clauses of the
knowledge base. For example, consider the following piece of knowledge and