Seismic risk analysis using Bayesian belief networks - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 185

Table 7.5 Performance assessment of different models for liquefaction
classifi cation

Search
algorithm

TP
rate

FP
rate

ROC
area Class Precision Recall F-measure

K2 Yes 0.875 0.333 0.805 0.875 0.839 0.884

No 0.667 0.125 0.772 0.667 0.715 0.884

Hill
climbing

Yes 0.933 0.333 0.815 0.933 0.870 0.894

No 0.667 0.067 0.863 0.667 0.752 0.894

TAN Yes 0.894 0.333 0.809 0.894 0.849 0.866

No 0.667 0.106 0.800 0.667 0.727 0.866

Tabu
search

Yes 0.913 0.364 0.798 0.913 0.852 0.875

No 0.636 0.087 0.824 0.636 0.718 0.875

Fixed-
structure
result

Yes 0.913 0.303 0.826 0.913 0.868 0.887

No 0.697 0.087 0.836 0.697 0.760 0.887

TP = true positive, FP = false positive, ROC = receive operating characteristic.

Chan et al. (1998) developed a general method of estimating earthquake
loss by considering gross domestic product (GDP). In comparison with
conventional approaches, this method requires less information and enables
loss assessment for any region with readily available GDP and population
data. Yong et al. (2002) utilized this method for the seismic hazard and loss
estimation in Central America. Balassanian et al. (1999) have developed a
similar method by considering building vulnerability, area of vulnerability,
and number of inhabitants. Carren˜o et al. (2007) applied a complex model,
with a wide array of base inputs, such as population density and hospital
beds, to evaluate the seismic risk for Bogota and Barcelona. This model
used a hierarchical structure with the base inputs that encompassed both
'hard' and 'soft' aspects contributing to seismic risk. More recently, Barbat
et al. (2009) compared such a holistic approach with a more traditional
vulnerability index and capacity spectrum, and discovered the potential
robustness of such a model and its ability to easily identify the most relevant
aspects of the urban seismic risk index. However, these models apply largely
to districts within city boundaries. Although this is invaluable information
for planners, it cannot quickly highlight which cities within a region require
such detailed analysis. The use of a hierarchical structure model to prioritize
seismically high-risk cities is gaining popularity (e.g. Roberts et al. 2009).
Davidson and Shah (1997) developed an earthquake disaster risk index
(EDRI) to assess seismic risk between urban cities. The EDRI has been
applied in many case studies (e.g. Sica et al. 2002; Zobin and Ventura-
Ramirez 2004; Cockburn and Tesfamariam 2012).