Evolutionary polynomial regression algorithm combined with robust bayesian regression. (May 2022)

• Record Type: Journal Article
• Authors: Marasco, Sebastiano; Marano, Giuseppe Carlo; Cimellaro, Gian Paolo
• Is Part Of: Advances in engineering software. Volume 167(2022)
• Publication Date: 2022-05
• Subjects: Evolutionary polynomial regression -- Robust Bayesian regression -- Machine learning -- Student-t distribution -- Shear strength; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2022.103101
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Novel technique combining Evolutionary Polynomial algorithms with Robust Bayesian Regression. Reducing the effects of anomalous data by using a disturbance model based on a Student- t distribution. The proposed technique is applied to a dataset consisting of experimental shear strength of RC beams without stirrups. Comparisons with current formulations have shown the highest accuracy for all the investigated levels of complexity. Abstract: Recent developments in the fields of scientific computation and Machine Learning (ML) techniques have led to a proliferation of algorithms capable of interpreting data and predict results. Among the others, the Evolutionary Polynomial Regression (EPR) has gained particular attention for many engineering applications. This paper presents a novel robust Evolutionary Polynomial Bayesian Regression (EPBR) algorithm. The optimal polynomial structure is selected using GAs. The model parameters are assumed as random variables whose posterior distributions are assessed by a robust Bayesian regression algorithm. To reduce the effects of the outliers, the model disturbance is described by a Student- t distribution whose degrees of freedoms are sampled from an objective prior probability density function. The proposed technique is applied to a dataset consisting of experimental shear strength values related to Reinforced Concrete (RC) beams without stirrups. The optimal EPBR model is compared with different experimental and design formulations to emphasize its accuracy and consistency.