Akademik Veri Yönetim Sistemi
Natural Hazards, cilt.122, sa.1, 2026 (SCI-Expanded, Scopus)
Earthquake record selection is crucial to ensure accurate representation of seismic demands on structures under seismic motions. Varied structural responses are obtained due to the vibrations from different earthquake types associated with fault proximity which is classified as near-fault (NF) and far-fault (FF). Within the near-fault category, motions are further classified into pulse-like and no-pulse types. However, there is no strict definition for the distinction among these earthquake types, despite the significant differences in the resulting structural responses to these vibrations. In this paper, earthquakes are classified based on fault proximity and pulse characteristics by using three machine learning methods, artificial neural networks (ANN), convolutional neural networks (CNN) and random forest (RF) algorithms. Three sets of earthquake data, each with 20 records, representing far-fault (FF), near-fault no-pulse (NF-NP) and near-fault pulse-like (NF-PL) characteristics, are utilized. For each method, models were carefully developed and trained using these datasets to capture the unique seismic features of each category and classify them to the associated set. The models corresponding to three machine learning methods were evaluated using key metrics, such as mean squared error, correlation coefficients and accuracy rates to determine their effectiveness in distinguishing between the fault proximity and pulse types of the earthquakes. Additional earthquakes are also tested to verify the models. The results not only highlight the capabilities of each model and their potential for contributing to improved seismic analysis and building safety assessments but also underline the superior performance of CNN to detect the fault proximity and pulse type.