Machine learning-aided time and cost overrun prediction in construction projects: application of artificial neural network

This research provides evidence of well-established evaluation frameworks for predicting time and cost overruns. There have been several attempts to reduce this issue, but these overruns still harm the construction industry worldwide. To create hyper-parameter-optimized predictive models, the numerical data was primarily used to train a specific type of machine learning algorithm known as an artificial neural network (ANN). In addition, Tabu Search was used to fine-tune the neural networks. Tabu Search returned a precision of 68.15826416015625, a recall of 80.6761884689331, and an accuracy of 92.19601929187775. The F1 score was 71.0405363922119. To better understand the occurrence of cost and schedule overruns, the study addressed in this article analyses 191 construction projects completed in the Hashemite Kingdom of Jordan between 2011 and 2021. The results showed that the R2 value was 0.938478 and the mean absolute error (MAE) was 0.057 for the data on cost overruns, while it was 0.9385 and the MAE was 21.7090 for the data on time overruns. Using tabular search optimization, the ANN model achieves better results than the baseline models. In sum, the neural network with tabular search optimization has proven to be a practical tool to employ in the early design phase, when there is typically a sparse or insufficient data set to conduct a cost and time overrun analysis. This approach has the potential to produce more reliable results and reduce the margin of error in estimates. © 2023, The Author(s), under exclusive licence to Springer Nature Switzerland AG.