Performance of artificial neural network system in prediction issues of earthquake engineering

Prediction is one of the most important issues of earthquake engineering. Empirical predictive relations are commonly played as basic rule in seismic hazard analysis. Such relations are generally expressed as mathematical functions connecting a strong motion parameter to the parameters characterising the earthquake source, the propagation path distance and the local site conditions. Regression analysis has been widely used among other analytical methods with different techniques during the past few decades, e.g. Gutenberg & Richter (1956), McGuire (1978), Joyner & Boore (1981), and Molas & Yamazaki (1995), etc.. Artificial neural networks were first applied to prediction issues of earthquake engineering by Emami et al. (1996). The performance of this advanced system in various aspects of prediction of ground motion parameters is discussed and compared with traditional procedures throughout this paper.