An accurate yet simple estimate of the retrofit cost plays an important role in the decision-making process of retrofitting existing buildings. Fung et al. (2018a) develop a predictive model to estimate seismic retrofit costs as a function of building characteristics such as building area, building age, and building model type. However, in practice, a decision maker may not have access to the full set of building characteristics required for estimating the retrofit cost, especially when dealing with a portfolio of buildings. Certain characteristics (e.g., building area) might be more readily available or easier to obtain than others (e.g., building type). This paper considers the tradeoff, in terms of prediction error, from not using all of the building characteristics necessary for prediction. The results show that excluding certain characteristics from prediction, such as building type, lead to negligible increases in prediction error. The paper also finds the minimal set of building characteristics needed to approximate the accuracy of the model that uses the full set of building characteristics. Findings of this study will help decisions makers to estimate retrofit costs without having to spend additional time and money to collect the full set of data on the building portfolio.
Proceedings of the 2nd Annual Conference on Natural Hazards & Infrastructure
June 23-November 26, 2019
2nd International Conference on Natural Hazards & Infrastructure
seismic retrofit, cost prediction, glm, regularization, variable selection