In this paper, we examined the performance of four fuzzy
rule generation methods that could generate fuzzy if-then
rules directly from training patterns with no timeconsuming
tuning procedures. In the first approach, a
single fuzzy if-then rule was generated for each class using
the mean and the standard deviation of attribute values. In
the second approach, a single fuzzy if-then rule was
generated for each class using the histogram of attribute
values. The third approach generated fuzzy if-then rules by
homogeneously partitioning each attribute. Thus, a pattern
space was partitioned into a simple fuzzy grid. The
information about attribute values was not used for
specifying the membership function of each antecedent
fuzzy set. The local information of training patterns was
utilized when the consequent class and the certainty grade
were specified. The last approach was a modified version
of the simple fuzzy grid approach.
As illustrated in Table 2, simple grid approach gave the
best performance overall while the mean and standard
deviation approach also performed reasonably well.
It may be noted that a single fuzzy if-then rule for each
class is not always sufficient for real-world pattern
classification problems. While each approach is very
simple and has some drawbacks as discussed above, fuzzy
rule-based systems have high classification ability as
shown in this paper. The performance of fuzzy rule based
systems can be further improved by feature selection and
optimizing the rule selection and various rule parameters.