A hybrid SEM-ANN model for predicting overall rework impact on the performance of bridge construction projects

Abstract

Determining the factors that cause rework in bridge is necessary to assess the impact of rework on a cost, which have been increasing in recent times. This research aims to evaluate the impact of rework on cost. In contrast to prior rework research, which focused heavily on structural equation modelling (SEM) analysis; this study uses a hybrid learning analytic approach based on the acceptance of a hybrid approach where the outputs of the second-order structural equation model are used as inputs to the artificial neural network, which derived from deep learning. Through the means of a questionnaire survey, 260 experts have participated in the study. The findings indicated that all rework factors have significant positive impacts on cost. There were 33 reasons for rework that affect the project performance. Using Delphi technology, these variables were reduced to 27. The final SEM model adapts 21 factors across five groups for reworking reasons. Wherein, the initial model RMSEA was 0.071, it became 0.057 in the final SEM model. The proposed hybrid model supports the relevance of the construction process-related factors in reducing rework on the bridge project. In comparison to the predicted value of the rework calculated by the hybrid model, the error rate was within 10 %. In practise, the findings will allow decision-makers and practitioners on construction sites to determine which aspects should be prioritised above others and design their policies appropriately. Methodologically, this study determines the deep ANN architecture's capabilities in finding non-linear correlations among the elements in the theoretical model.