[3Rin4] Interactive 1

[3Rin4-03] Applicability of features to ground motion evaluation models utilizing machine learning

〇Atsuko Oana¹, Toru Ishii¹, Kensuke Wada¹ (1.Shimizu Corporation)

Keywords:Machine learning, Ground motion evaluation, Kanto region, Features

The aim of this paper is to explore the applicability of features which should be considered in seismic ground motion evaluation models utilizing machine learning. First, in order to obtain clues for selecting features, we create a ground motion evaluation model using more features than the general parameters used in the conventional ground motion prediction equations (GMPEs). Then we evaluate each feature impact on the target variable. Next, we attempt to interpret physically the evaluation results suggested by each feature impact and the correlations between the features. Finally, we compare the evaluation accuracy and characteristics of the models using the extracted features and those equivalent to the parameters of the conventional GMPEs. This result is expected to provide a new insight for constructing more accurate ground motion evaluation models in the future.