Mutual information-based feature selection for estimating an appropriate vector-valued seismic intensity measure

Abstract

Performance-based seismic design requires the use of probabilistic methods to predict the structural response to ground motion. This approach relies on the seismic demand model, which links intensity measures (IMs) with damage measures to evaluate structural response. It's important to choose the appropriate IM for a correct evaluation of performance. According to new research, vector-valued IMs with multiple scalar components may help make predictions more accurate by lowering dispersion and better capturing important ground motion characteristics. Despite this, it is still challenging to pick the appropriate vector-valued IM because it has to meet both efficiency and sufficiency criteria. This paper presents a novel approach based on information theory to tackle this challenge. The method involves picking the first and second components of the vector-valued information measure based on mutual information, conditional mutual information, and redundancy analysis. This way, the most useful information is found while redundancy is kept to a minimum, and the correlation between the vector-valued information measure components is taken into account. The method is tested on two reinforced concrete structures that are four and eight stories, using a set of 32 candidate scalar IMs and 60 ground motion records. The findings illustrate the efficacy of this method in identifying an appropriate vector-valued IM.