Structured versus direct-mapping approaches to empirical modeling of civil engineering problems

Empirical modeling and its associated methodologies are widely adopted in the development process of prediction systems for solving construction engineering problems. Most current empirical models employ a direct mapping architecture with only one mapping step between input and output variables. This property limits the model extendibility by preventing the model from solving some extended versions of the civil engineering problem it is originally developed to solve. In addition, the data size used to train and test the model grow exponentially when new independent variables are introduced. This study aims to develop a new approach to circumvent the above mentioned issues in the application of empirical models to civil and construction problems. The study investigated the performances of several direct mapping models in solving truck type classification based on weigh-in-motion data and resolve this problem using a newly developed empirical modeling schema called structured empirical modelling approach (SEMA). The innovation is inspired by the adoption of intermediate features in the deep machine learning field. Proposed benefits of SEMA modeling includes, but is not limited to, improved model prediction accuracy, improved model flexibility in training and testing, reduced training data requirements and potential extendibility to solve similar construction engineering problems.