Earthquake induced damage classification for reinforced concrete buildings

Abstract

Seismic risk assessment of reinforced concrete buildings needs consideration of seismic hazard, building vulnerability and consequence of failure. Different statistical methods are proposed to discern vulnerable buildings for retrofit prioritization. This paper utilized reported seismic induced damage data and illustrated eight different statistical damage classification techniques, naive Bayes, k-nearest-neighbor (kNN), Fisher's linear discriminant analysis (FLDA), partial least squares discriminant analysis (PLSDA), multilayer perceptron neural networks (MLP-NN), classification tree (CT), support vector machine (SVM), and random forest (RF). Six building performance modifiers were considered in this study for damage classification: number of stories above the ground level (N), soft story index (SSI), overhang ratio (OHR), minimum normalized lateral stiffness index (MNLSTFI), minimum normalized lateral strength index (MNLSI) and normalized redundancy score (NRS). The results demonstrate the feasibility and effectiveness of the selected statistical approaches to classify the damage of concrete buildings.