y

g

y

g

omes.

the Dayhoff matrix, the learning was terminated at the 67^th cycle.

ss measurement and the AUC value of the top rule were 0.84 and

pectively at this stage. Figure 8.26 shows the evolving AUC

nd fitness measurements along with the learning cycle. In two

h the maximum and the median measurements were included for

C and fitness. Because the best rule was maintained unless it was

by an even better one, the maximum AUC curve and the

m fitness measurement curve had almost no oscillation. However,

an measurements had oscillations. This was not a surprise because

breeding rules had their performance varying all the time.

r, the periodically great oscillations were due to the introduced

n process.

(a) (b)

he Dayhoff GP models. (a) The AUC value curve. (b) The fitness measurement

RPN chromosome of the top rule which had the greatest fitness

ment was (dRaUcYeUcG(dWcY)&)+. Figure 8.27 shows its tree

The mathematic expression of this rule is shown below,

ሺܠ, ܚሻൌሼminሼmaxሼߨሺݔଷ, Yሻ, ߨሺݔସ, Wሻሽ,ߨሺݔଵ, Uሻ, ߨሺݔଷ, G

/Yሻ, ߨሺݔସ, Rሻ, ߨሺݔହ, ܷሻሽሽ