Investigating the Capabilities of Ensemble Machine Learning Model in Identifying Near-Fault Pulse-Like Ground Motions

In recent years, the detection and analysis of near-fault pulse-like ground motions have become increasingly crucial due to their potential to cause significant structural damage during seismic events. These motions, characterized by their unique directivity or fling step effects, pose a substantial challenge in earthquake engineering. Traditionally, wavelet decomposition has been employed to dissect the ground motion time history, extracting a pulse signal to aid in the classification of seismic events. However, this method demands extensive computational resources and complex calculations. This study applies various ensemble machine learning models, such as random forests, gradient boosting machines, and extreme gradient boosting, for the identification and characterization of pulse-like ground motions. The research integrates multiple ensemble techniques to develop a robust and efficient framework for seismic data analysis. These models offer superior performance in handling large datasets, noise reduction, and feature extraction, making them ideal for this application. The study evaluates the effectiveness of these ensemble models in comparison to traditional methods, focusing on their ability to manage the unique attributes of pulse-like ground motions. The results demonstrate that ensemble machine learning models not only enhance the accuracy of identification but also significantly improve the prediction of pulse periods, achieving an accuracy of 93.4% for classification and a mean absolute error of 1.52 seconds for regression. This research contributes to the field of earthquake engineering by providing a more reliable and computationally efficient tool for the analysis of near-fault pulse-like ground motions, ultimately aiding in the development of better earthquake-resistant structures.