A comparative study of Artificial Intelligence methods for project duration forecasting

• We propose 5 AI methods for predicting a project’s duration.
• A methodology including PCA, cross-validation and grid search is presented.
• A large computational experiment shows the good performance of the AI methods.
• A sensitivity analysis reveals the weakness of the proposed methods.

This paper presents five Artificial Intelligence (AI) methods to predict the final duration of a project. A methodology that involves Monte Carlo simulation, Principal Component Analysis and cross-validation is proposed and can be applied by academics and practitioners. The performance of the AI methods is assessed by means of a large and topologically diverse dataset and is benchmarked against the best performing Earned Value Management/Earned Schedule (EVM/ES) methods. The results show that the AI methods outperform the EVM/ES methods if the training and test sets are at least similar to one another. Additionally, the AI methods report excellent early and mid-stage forecasting results. A robustness experiment gradually increases the discrepancy between the training and test sets and demonstrates the limitations of the newly proposed AI methods.