Page 230 - Data Science Algorithms in a Week

214 Abdulrahman Albar, Ahmad Elshennawy, Mohammed Basingab et al.

assigned boundaries to create evenly distributed classes on the crowding axis, and
similarly to subsystem III and I, the degree of membership is equivalent to 1 among the
two classes existing at any given point. Only at the points 0, 25, 50, 75, and 100, do the
five respective classes individually obtain full degrees of membership.

Figure 19: Membership function of patient boarding. Figure 20: Membership function of crowding.

Results of Expert Evaluation

This section presents the results of the fuzzy rule base development and the experts’
consensus rate. The fuzzy rule base assessments are divided by subsystem, with
subsystem I producing 120 rules assessments, and subsystem II, III, and IV producing 90,
360, and 800 rule assessments, respectively, for a total of 1370 assessments obtained.
After reaching consensus, the final version of the fuzzy rules is listed in this section.
Table 7 details the results from the expert assessment of the fuzzy rules from
subsystem I. This table consists of 12 columns, beginning with the rule code, followed by
ten expert evaluations, and ending with consensus status. Below the table is a legend
comprising five linguistic classes which are color-coded. In this subsystem, two fuzzy
rules reached full consensus (100%); FLS1-11, and FLS1-12. Two rules achieved 90%
consensus: FLS1-05, and FLS1-06; four reached 80%: FLS1-01, FLS1-04, FLS1-07, and
FLS1-08; one rule reached 70% consensus: FLS1-03, and three reached 60% consensus:
FLS1-02, FLS1-09, and FLS1-10. The average consensus rate for this subsystem’s rule
assessments is 79%. Seven of the twelve evaluated rules received assessments across
only two linguistic classes, while two were assessed across three linguistic classes, and
only one received assessments exceeding more than three types of linguistic assessment.
Most the data in this subsystem is centralized around two linguistic classes. Regarding
the frequency of linguistic class use, “medium” was most frequently used to assess rules,
with 42 uses, while “high”, “low”, “very high”, and “very low” were used 30, 21, 15, and
12 times, respectively.
All of the fuzzy rule statements for subsystem I (Appendix A), after consensus, are
listed according to their rule number. This final version of the rules will be stored in the
fuzzy rule base of the knowledge base to fuel the fuzzy inference engine.