Three data sets are used for the experiments reported in the next section. One artificial data sets is useful for this study because it has local extrema for Jm.
The artificial
data set having multiple local extrema is a one-dimensional
data set (single feature data set) consisting
of one feature, the output y of the non-linear equation y=(1+
x-21 +x-1.52)2, where 
.This equation is taken from
[12], where 6 classes were found to provide a useful grouping
in the development of fuzzy rules to model the output of the
equation over a set of 50 distinct values. The same 50 y values were
used in this study.
The other two data sets are the Iris data and multiple sclerosis data [9]. Iris consists of 150 patterns belonging to 3 species of Iris, with 50 patterns in each class. Each of the patterns is described by 4 real-valued features: petal-length, petal-width, sepal-length, and sepal-width.
The multiple sclerosis (MS) data consists of 2 classes each described by 5 measurements (features) of a patient. The classes are MS (29 examples) and non-MS (69 examples). The first feature is age, and measurements connected with two different visual stimuli (S1 and S2) provide the other 4 features. These four features are the sums and absolute differences of the responses of the stimuli as observed in the left and right eyes, respectively. Letting L stand for left eye response and R stand for right eye response, these features are (S1L + S2L), (|S1L - S2L|), (S2L + S2R), and (|S2L - S2R|).