Parametrized division of exposure zone for marine reinforced concrete structures with a multi-class Boosting method

•A multi-class Boosting method is proposed for the multi-class imbalanced problem.•Parametrized division of exposure zone is achieved using RMS, Cr, Cs, and Dcl, etc.•Field chloride profile data with time span of 18 years validates proposed method.•The distribution of exposure zone satisfies the Weibull distribution. The analysis of marine reinforced concrete structures using chloride profile data is a commonly used exposure zone classification method. However, chloride profile data is multi-class, unbalanced and non-parametric, which makes it difficult for the commonly used machine-learning methods to construct an appropriate classification model. To solve this problem, chloride profile is parametrized by the minimum redundancy maximum relevance algorithm and a multi-class Boosting method using F-measure as inductive bias indicator to evaluate the weight of base classifiers is put forward. The method is based on field test data of the chloride profile over a period of 18 years in Hangzhou Bay, China. The method outperforms the original Boosting method with an average F-measure improvement of 6.2 %. The results show that parametric partitioning of the exposure zone is achieved and the distribution of the exposure zone satisfies the Weibull distribution.