A generalized artificial neural network (ANN) model for predicting construction project durations based on cost and geolocation data
Mehmet Sena Kaşka, Isik Ates Kiral, Anıl Niş, Semih Caglayan, Sevilay DemirkesenPurpose
Reliable construction project duration estimation remains a persistent challenge due to the complex interactions among project characteristics and the diversity of construction environments. While previous prediction models have frequently relied on project-specific variables and geographically limited datasets, their applicability across different regions remains uncertain. This study aims to develop a generalized artificial neural network (ANN) model capable of predicting construction project durations using universally available project information-cost and geographical location-to provide a practical and transferable decision-support tool for early-stage project planning.
Design/methodology/approach
A heterogeneous dataset comprising 545 construction projects from multiple countries was used to develop and validate an ANN model. Following data preprocessing and standardization, project cost and geographical location were employed as input variables to predict project duration. The dataset was divided into training, validation and test subsets to evaluate predictive accuracy, generalization capability and model robustness across diverse construction environments.
Findings
The proposed ANN model achieved high R2 scores on all subsets (training: 0.8752, validation: 0.8421, test: 0.8600), indicating a robust capacity to capture underlying nonlinear relationships between input variables and project duration. Visual analyses, including performance curves, error histograms and training diagnostics, further supported the model's stability and low risk of overfitting. The results not only align with but also extend previous findings in the field by demonstrating superior generalization over geographically diverse data.
Originality/value
The study contributes both theoretically – by reinforcing ANN's applicability in construction informatics – and practically – by offering a decision-support tool for early-stage planning when historical data may be limited. Recommendations for future work include model explainability enhancements and integration into digital project management platforms.