Exploring the best ANN model based on four paradigms to predict delay and cost overrun percentages of highway projects

Abstract

This study explores the best models in predicting delay and cost overrun percentages (PDCOP) for highway projects depending on four ANNs-based paradigms: principal component analysis (PCA), modular neural network, REF/GRNN/PNN network and time-lag recurrent network. The best model among 28 developed for predicting % delay based on modular neural network paradigm with: PCA as an input projection algorithm, Online training as weight update, Sigmoid Axon as transfer function, Momentum (0.7) as learning rule for hidden and output layers gives % prediction with Mean Absolute Percentage Error (MAPE) equals to 39.8%. Whereas, the best model among 28 developed for % cost overrun, based on PCA paradigm with the same characteristics as best model developed for % delay and learning rate (0.01), has MAPE equals to 25.4%. Furthermore, the best proposed models for PDCOP outperform the previously recently models in literature. MAPE equals to 39.8% for best proposed model for predicting % delay versus 53.68 and 57.68% for linear regression and statistical fuzzy models proposed in literature. Whereas, MAPE equals to 25.4% for best proposed model for predicting % cost overrun versus 30.42 and 40.37% for models in literature, respectively.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 30 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.