Duration prediction of Chilean strong motion data using machine learning. (August 2021)

• Record Type: Journal Article
• Authors: Chanda, Sarit; Raghucharan, M.C.; Karthik Reddy, K.S.K.; Chaudhari, Vasudeo; Somala, Surendra Nadh
• Is Part Of: Journal of South American earth sciences. Volume 109(2021)
• Publication Date: 2021-08
• Subjects: Duration -- Significant-duration -- Inslab -- Classifiers -- Machine learning algorithms; Geology -- Latin America -- Periodicals; Earth sciences -- Latin America -- Periodicals; Geology -- Antarctica -- Periodicals; Earth sciences -- Antarctica -- Periodicals; Geology -- Caribbean Area -- Periodicals; Earth sciences -- Caribbean Area -- Periodicals; Géologie -- Amérique latine -- Périodiques; Sciences de la terre -- Amérique latine -- Périodiques; Géologie -- Antarctique -- Périodiques; Sciences de la terre -- Antarctique -- Périodiques; Géologie -- Caraïbes (Région) -- Périodiques; Sciences de la terre -- Caraïbes (Région) -- Périodiques; Earth sciences; Geology; Antarctica; Caribbean Area; Latin America; Periodicals; 550.5
• DOI: 10.1016/j.jsames.2021.103253
• Languages: English
• ISSNs: 0895-9811

Abstract

Abstract: Chile is rocked by inslab, interface as well as crustal events. Duration estimates based on Chilean strong motion flatfile is used to predict total duration as well as significant-duration. We use six different machine learning algorithms k-nearest neighbours, support vector machine, Random forest, Neural network, AdaBoost, decision tree and estimate the accuracies of prediction for each component (EW, NS, Z) of ground motion for different tectonic environments. The estimates of duration using machine learning are found to be quite accurate and the best performing machine learning algorithm in prediction of the total duration and the significant-duration are highlighted. Highlights: Machine learning approaches have been tried out on Chilean strong motion to predict earthquake duration. Ensemble machine learning algorithms have also been used. Results are partitioned by tectonics (inslab, interface, crustal). Best performing machine learning algorithm is highlighted in each case.