The structure and parameters of artificial neural networks (ANNs) are typically selected based on experience. This study aims to present a dynamic optimization model using metaheuristic algorithms in combination with ANN to achieve more reliable labor productivity predictions in construction project activities.
This study focuses on optimizing and selecting the best metaheuristic algorithm from a set of candidates in conjunction with ANN. The goal is to intelligently attain the optimal combined model for weights and biases by minimizing mean squared error and ensuring an appropriate convergence process. This approach facilitates more accurate predictions of the target variable.
A case study was conducted to predict labor productivity in a construction project. Seven metaheuristic algorithms – Salp Swarm Algorithm, Particle Swarm Optimization, Grey Wolf Optimizer, Whale Optimization Algorithm, Bat Algorithm, Firefly Algorithm and Sine Cosine Algorithm – were used as candidates for integration with ANN to forecast workshop labor productivity. The findings of this study indicated that, in alignment with the case study data, WOA-based training performed better than other candidates in the intelligent system, providing more accurate results in terms of both accuracy and convergence.
This flexible hybrid dynamic model enhances the reliability of predictions related to activity data sets by automatically identifying the best optimizer. It can adapt to various workshop conditions and serves as a valuable tool for managers to understand and implement timely management strategies and techniques for improving productivity.
The proposed flexible hybrid intelligent model increases the accuracy of labor productivity estimates tailored to diverse activities in construction projects. It is repeatable and adaptable to various workshop conditions. Compared to other candidate combinations, the proposed model intelligently discovers the optimal algorithm, surpassing conventional methods in terms of accuracy, convergence, performance evaluation and error reduction.
