Comparative Study on Data Mining Methods in Structural Reliability Prediction

Willy Husada (Institut Teknologi Sepuluh Nopember, Surabaya)
I-Tung Yang (National Taiwan University of Science and Technology (NTUST))
Tri Joko Wahyu (Institut Teknologi Sepuluh Nopember, Surabaya)

IPTEK Journal of Proceedings Series
No 1 (2015): 1st International Seminar on Science and Technology (ISST) 2015

The goal of reliability-based design optimization (RBDO) is to find the optimal structure design with minimum cost subjected to maximum failure probability limit. Since failure probability is usually small, it takes a large amount of computation time for accurate estimation in reliability analysis. Surrogate models usually created to replace the time-consuming reliability analysis. In this empirical study, we use several data mining methods with focus on classification and regression tree (CART), artificial neural network (ANN) and support vector machine (SVM) method to create the surrogate models on a empirical benchmark case study. We aim to find the best data mining method in predicting the failure probability which divided into two parts: classification and regression. The main findings of this study is that CART method performed better than ANN and SVM in both classification and regression. Support vector machine (SVM) method is the worst in both cases.