Study on Reworks for Commercial Complex Projects Based on Deep Learning Neural Network

Abstract Commercial complexes have emerged as the main form in the modern urban development, the implementation process of which is characterized by diverse professional teams, multiple levels of construction, and more stakeholders. However, reworks and associated costs in commercial complex projects have not been addressed. This study focuses on reworks of the commercial complex projects including design and construction. The actual commercial complex projects with data of rework costs are based to identify controllable factors contributing to reworks and collect rework dataset including scores of the factors and rework costs. A deep learning neural network (DLNN) in multi-layer perceptron (MLP) form is utilized to predict rework costs, and is justified by comparing with other neural network applications. A case commercial complex project is applied to demonstrate the proposed MLP-based DLNN in predicting rework cost and analyze the importance of rework factors with SHAP values. With investigation into rework factors and prediction of rework costs, this study enables to achieve management and control measures on reworks in the design and construction processes of commercial complexes, thus promoting cost-effectiveness and sustainability in modern urban construction.