Exploring a recurrent neural network (RNN) earned value based model for predicting lost labour value in construction projects

Abstract

Abstract This study explores the application of deep learning techniques in forecasting labour losses in construction projects. By leveraging the predictive capabilities of ManHour Earned and Estimated ManHour values, this research aims to develop an accurate and reliable model for labour loss forecasting. Utilizing a dataset comprising 105 work items from industry practitioners who employed the Leonard and Moselhi methodology, this study investigates the efficacy of a Recurrent Neural Network (RNN) architecture with Long Short-Term Memory (LSTM) cells. The model's performance is optimized through backpropagation and hyperparameter tuning, focusing on key parameters such as learning rate, batch size, dropout rate, and number of epochs. To ensure the model's generalizability and mitigate overfitting, regularization strategies and performance monitoring are employed. The predictive potential of the model is evaluated through integration into a real-time sequence production job and deployment into an enabled environment. Results demonstrate the model's capacity for accurate labour loss forecasting, enabling proactive measures to enhance productivity and minimize unnecessary claims. Although this study focuses on building projects, its findings have broader implications for process and manufacturing engineering, where productivity losses and associated labour losses are significant concerns. The practical application of this research lies in its potential to assess and counter labour loss claims, disputes, warrants, and demands in construction projects, ultimately contributing to improved project management and reduced financial losses.

Keywords

• Deep learning
• Lost labour forecasting
• Recurrent neural network
• Long short-term memory
• Hyperparameter optimization