A locally weighted machine learning model for generalized prediction of drift capacity in seismic vulnerability assessments

Abstract

Drift capacity of reinforced concrete (RC) columns is an important indicator to quantify the seismic vulnerability of RC frame buildings; however, it is challenging to accurately predict this value as the nonlinear behavior can vary greatly by column type. This article proposes a novel, local machine learning (ML) model, called locally weighted least squares support vector machines for regression (LWLS‐SVMR), which integrates LS‐SVMR and locally weighted training criteria to enhance and generalize the prediction of the drift capacity of RC columns, regardless of the type. A database of 160 circular RC columns covering flexure‐, shear‐, and flexure–shear‐critical specimens was developed to train and test the proposed LWLS‐SVMR. The proposed LWLS‐SVMR was validated by comparison with popular existing global and local learning approaches as well as a traditional empirical equation, and the results demonstrated that the proposed LWLS‐SVMR is superior to all other approaches and thus, is a promising artificial intelligence technique for enhancing the prediction of drift capacity, universally across RC flexure‐, shear‐, and flexure–shear‐critical columns. The LWLS‐SVMR exhibits capabilities which may yield it a feasible approach to predict complex, nonlinear behavior in a broad‐spectrum manner.

The promise of implementing machine learning in earthquake engineering: A state-of-the-art review

“…Recently, ML has been utilized to predict the displacement capacity of RC columns, where input data that are available for shear-critical columns are much fewer than those for flexure-critical columns. Consequently, ML models for shear-critical columns do not perform as well as those for flexure failures (Luo and Paal, 2019). Note that such data quantity issues cannot be easily tackled by switching or developing a more advanced ML model (Domingos, 2012).…”

“…Recently, a multi-output least-squares support vector machine (MLS-SVMR) algorithm was implemented by Luo and Paal (2018) to construct bilinear force-displacement constitutive relationships for RC columns. A similar study has been conducted by Luo and Paal (2019) to predict the drift capacity of RC columns using a locally weighted least square SVR approach. Also, a group of six ML algorithms has been utilized by Huang and Burton (2019) to classify the in-plane failure modes of RC frame structures with masonry infill panels, and 114 test results from infill frame specimens were investigated.…”

Reducing the effect of sample bias for small data sets with double‐weighted support vector transfer regression

“…Because only the drift capacity (i.e., of ductile columns will be used to constitute the source domain data (see Sections 5.2.2 and 5.2.3 for more detailed information), the statistical properties for the drift capacity of ductile columns are given. More detailed information for the rectangular and circular RC column data sets can be found in Luo and Paal (2018, 2019), respectively.…”

“…One challenge, which significantly affects their performance, is how to reduce the negative effect induced by sample bias of small data sets, specifically in regression scenarios. This is because regression‐based ML techniques usually require a large, high‐quality training data set to adaptively fit the data and form an accurate, robust model for prediction (Ahangar‐Asr, Faramarzi, Javadi, & Giustolisi, 2011; Aminian, Javid, Asghari, Gandomi, & Esmaeili, 2011; Cheng & Cao, 2014; Chou & Pham, 2015; Gandomi, Mohammadzadeh, Pérez‐Ordóñez, & Alavi, 2014; Jeon, Shafieezadeh, & DesRoches, 2014; Luo & Paal, 2018, 2019; Pal & Deswal, 2011; Rafiei & Adeli, 2016, 2018; Rafiei, Khushefati, Demirboga, & Adeli, 2017; Yuen, Ortiz, & Huang, 2016). Typically, the sample points in a training data set can reasonably represent the distribution of a target domain.…”

Deep learning techniques for recommender systems based on collaborative filtering

“…Those methods rely on non-linear optimization techniques to learn model parameters, which make them substantially more robust for data representation, and their high generalization strength also facilitates dealing efficiently with raw or non-structured data. In this way, deep neural networks are able to achieve deeper abstractions of data, effectively F I G U R E 1 -An example of rating matrix demonstrating considerable advances in a variety of tasks like detection (Ansari et al, 2019;Bang et al, 2019;Maeda et al, 2018;Maeda et al, 2019;Zhang, Cheng, & Ren, 2019), prediction (Luo & Paal, 2019;Nguyen et al, 2019), clustering (Gao et al, 2019;Reyes & Ventura, 2019), and classification (LeCun et al, 2015;Maeda et al, 2018;Manzanera et al, 2019). Last but not least, we emphasize that many other classification techniques could be considered to cope with the problem of CF, such as the Enhanced Probabilistic Neural Networks (Ahmadlou & Adeli, 2010), Neural Dynamic Classification (Rafiei & Adeli, 2017), and the Finite Element Machine classifier (Pereira et al, 2020), Consequently, the study and development of deep learning techniques have facilitated important improvements in many computer science research areas, from which we quote object detection (Antoniades et al, 2018;Molina-Cabello et al, 2018;Vera-Olmos et al, 2018;Wang & Bai, 2018), speech recognition, computer vision Shen et al, 2019), and natural language processing (LeCun et al, 2015).…”