Seismic risk analysis using Bayesian belief networks - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 181

0.1 0.3 0.5 50 150 0.0 0.2 0.4

**** **** **** ****

M

0.541 0.432 0.575 0.0674 0.361

**** **** ***

a

0.0823

0.335 0.509 0.207

max

**** *** ****

s vo

0.929 0.224 0.340

**** ****

s¢ vo

0.268 0.308

****

q c

0.277

D

50

6.0 7.0 50 150 250 0 5 15 25

7.4 The correlation matrix of considered variables (liquefaction).

algorithms, K2, hill climbing, tree-augmented naïve Bayes (TAN), and Tabu
search (Bouckaert et al. 2011), were tested with the available dataset. A
brief description of these searching algorithms is provided in Table 7.3.
Detailed information is available in the corresponding references.
The BBN relationships are generated through the four search algorithms
and results are depicted in Fig. 7.5. The K2 had all input parameters at the
same level, whereas the hill climbing search algorithm identifi ed the rela-
tionship between M and a max . The T AN and Tabu search algorithms showed
counter-intuitive results in terms of the parameter dependence. For example,
both show that M is dependent on D 50 .
Based on the engineering judgment and known relationship between
different input parameters, a fi xed structure model is proposed as shown in
Fig. 7.6. The conditional probability tables are generated by the specifi c
searching algorithm with the available training data set. The classifi cation
results and performance measures of each algorithm are also summarized
in Tables 7.4 and 7.5, respectively. As shown in Table 7.4, the BBN model