Evaluation of parameters affecting earthquake damage by decision tree techniques

Abstract

Earthquake damages are assessed based on a holistic approach using structural as well as non-structural factors to model earthquake damage distributions with Decision Tree Techniques, using the Answer Tree program and the damage data from recent major earthquakes in Turkey. The damage dataset consists of approximately 9,400 buildings that were surveyed to evaluate the factors affecting building damage after Erzincan [1992], Dinar [1995], and Kocaeli [1999] earthquakes. The earthquake damage is defined as the dependent variable, while earthquake magnitude (M), intensity (I) in the city, peak ground acceleration (PGA) in each city, epicenter distance (ED), building types (BT), number of storeys (NS), presence of soft storey (SS), building position (BP) on the site, and site conditions (SC) are independent variables in the proposed model. The damage level (DL) was classified with respect to red, yellow, and green codes. The main purpose was (1) to identify the factors controlling building damage during earthquakes; (b) to determine the most significant factor; (c) to evaluate the effects of different factors for different earthquakes; (d) to develop damage distribution models for different subgroups based on the Decision Tree Techniques.