M‌A‌C‌H‌I‌N‌E L‌E‌A‌R‌N‌I‌N‌G-B‌A‌S‌E‌D R‌E‌G‌R‌E‌S‌S‌I‌O‌N M‌O‌D‌E‌L‌S F‌O‌R E‌S‌T‌I‌M‌A‌T‌I‌N‌G S‌E‌I‌S‌M‌I‌C R‌E‌T‌R‌O‌F‌I‌T C‌O‌S‌T O‌F M‌A‌S‌O‌N‌R‌Y B‌U‌I‌L‌D‌I‌N‌G‌S

Abstract

Retrofit actions are amongst the most commonly used measures for reducing the seismic vulnerability of buildings. For any given building, multiple seismic retrofit options are often available. Each option has specific requirements, cost, and performance. Estimating the cost of each candidate action is essential to the selection, planning, and implementation of seismic retrofit initiatives. Primary cost estimation plays a vital role in allocating budget for retrofit projects. Past studies used a variety of methods to develop cost estimation models. This research harnesses the capabilities of various regression models via modern machine learning methods for cost estimation. A dataset from 167 retrofit projects for masonry school buildings in Iran was used to develop models. Three main retrofit actions were implemented in the projects, namely Shotcrete, Steel belt, and Fiber reinforced polymer. Several regression methods including multiple linear regression, ridge regression, lasso regression, and also elastic net regression were applied to the dataset. The proposed framework comprised 12 models, which were attained by four regression methods on three retrofit actions. The cross-validation method was used for model evaluation in order to use all available data for training and testing. The model at the beginning of the development process contained all the probable effective parameters. Next, to increase the simplicity and accuracy of the models, a simple model reduction method was implemented. This model reduction method eliminated almost two-thirds of the parameters in the majority of basic models. Then, the candidate models were evaluated in terms of quantity and quality of prediction, heteroscedasticity, autocorrelation of residuals, and non-normality. This paper identifies the height of the building as the most influential parameter governing retrofit cost. Furthermore, lateral area of walls, footprint area, and added lateral strength are influential in the mentioned retrofit actions. This research contributes to enhancing the understanding of the factors, the effects, and the costs of the retrofit actions.

Keywords

• seismic retrofit
• cost estimation
• machine learning
• regression