An automated machine-learning-assisted stochastic-fuzzy multi-criteria decision making tool: Addressing record-to-record variability in seismic design

While uncertainty quantification (UQ) has served a prominent role in ensuring the safety of dynamical engineering systems, the lack of an integrated approach to handle the aleatory nature of ground motion records, a.k.a., record-to-record (RTR) variability, remains a bottleneck in seismic design. This paper presents a novel approach with two key features. First, on the multi-criteria decision-making (MCDM) front, a general-purpose collective decision support system is introduced by integrating Monte Carlo simulation, automated machine learning (AutoML), and a hybrid fuzzy-outranking MCDM technique. This allows for robust uncertainty capture at a group level without compromising computational efficiency—a departure from the traditional trade-off between MCDM competency in uncertainty handling and computational burden. Second, from an Earthquake Engineering perspective, the established AutoML-based community-level MCDM approach is combined with an efficient metamodel-aided reliability-based design optimization technique and a novel probabilistic-fuzzy seismic-design safety index. Conflicting criteria are classified into structural performance, design safety, and construction cost metrics. The results indicate that the developed intelligent seismic design paradigm properly captures uncertainties rooted in the RTR variability, input data, criterion weighting, and decision-makers’ preferences from a community-level standpoint. It also offers acceptable error metrics during the ranking procedure in a user-friendly environment, at a significantly reduced computational expenditure. Moreover, it can be concluded that there is no imperative need for developing complex decision tools for robust-to-uncertainty seismic design. The implementation of the proposed MCDM-based UQ framework has reduced the design cost of an existing dam by 17%.