The Exposition of Fuzzy Decision Trees and Their Application in Biology - Advancing Artificial Intelligence Through Biological Process Applications - page 391

namely TEMP and AGE. Again, three MFs are
used to define the linguistic terms which make up
the linguistic variable versions of these attributes,
see Figure 8.
The two sets of MFs presented in Figure 8,
along with the other nominal and ordinal condi-
tion attributes, WINTER, SEX and YEAR, can be
used to construct the associated FDT, which here
assesses the relevance of their association with
the duration of topor bouts. In this construction
process a necessary subsethood threshold value
of b = 0.72 was adopted, upon which a node is
deemed a leaf node.
The first stage of the FDT construction process
was the evaluation of the classification ambigu-
ity G ( E ) values associated with the five condi-
tion attributes describing the bats, found to be,
G (TEMP) = 0.519, G (AGE) = 0.733, G (WINTER)
= 1.078, G (SEX) = 1.068 and G (YEAR) = 1.068.
Inspection of these values shows, which range
from 0.519 to 1.078, the TEMP condition attri-
bute possesses the least classification ambiguity,
so forms the root node, it s noteworthy that the
three non-continuous attributes have the worst
associated levels of classification ambiguity. The
construction process can be continued in the same
way as previously described, the resultant FDT
is reported in Figure 9.
The reported FDT in Figure 9 is shown to have
three levels, indicating a maximum of two attri-
butes are used in the concomitant condition parts
of the fuzzy decision rules derived. Indeed, only
two attributes are used in the FDT shown, namely
TEMP and AGE, with seven fuzzy decision rules,
R1 , R2 , .., R7 , shown to be constructed. Of these
seven rules five of them have an associated truth
level, labelled with a *, that indicates they have
a truth level bellow that desired, since no further
augmentation of other condition attributes was
worthwhile. One of these rules is next explained
more fully, namely R7 , expressing it with the most
linguistic version of the rule;

'If average temperature is high and age is high then
the torpor bout is low with truth level 1.000.'

Inspection of the whole FDT produced indi-
cates ambient temperature was found to have a
negative association with torpor bout duration,
and these results suggest that older bats especially
have shorter torpor bouts at higher temperatures.
This result is consistent with predictions that
torpor lasts until a critical metabolic or water
imbalance is achieved. Because metabolism and
water loss are temperature-dependent, torpor bout
duration decreases with increased temperature
(Park et al ., 2000). In the original analyses by
Park et al . (2000), the influence of age on torpor
bout duration was not investigated, although it
was noted that there were no significant differ-
ences found between individuals. Since only 3 of
the 12 bats were older than one year, the effect of
age on torpor bouts shown here should be treated
with caution.
Further inference can be gauged from this FDT,
even though it is a small FDT with only a limited
insight into the problem, unlike the larger FDT
constructed to investigate the abalones. Perhaps
these latter results may suggest further condition
attributes should be found that may, through an
FDT analysis, elucidate further interpretability
into the problem.

FUTURE RESEARCH DIRECTIONS

Within a biological context, the rudiments of any
analysis are to achieve a greater understanding of
the issue being investigated. When analysing a
data set of objects, such as the abalones and bats
considered in this chapter, the precision of the
attributes describing them may not be certain,
as well as the a level of uncertainty with their
association to the decision problem in question.
The fuzzy decision tree (FDT) technique offers a
novel approach to the wide range of classification