Intentionally biased bootstrap aggregated artificial intelligence method for construction management prediction

Date

2025-9-1

Author

Çaydar, Tunahan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

6063
views

0
downloads

Cite This

Accurate prediction of fundamental construction project parameters such as: cost, duration and manpower requirements is critical to make adequate plans and management decisions for achieving successful construction projects, particularly during early stages when information is scarce. Traditional construction prediction methods rely on linear assumptions and expert judgement, failing to capture the non-linear relationships inherent in complex datasets. Machine learning techniques offer powerful tools for modelling such complexities; however, these techniques often exhibit overfitting, excessive sensitivity to noisy or skewed data and limited generalisation, in particular when trained with small-scale datasets, which is a significant limitation for the majority of construction prediction machine learning models. Ensemble learning, which aggregates predictions from multiple base models to enhance robustness and accuracy, has therefore gained prominence. This study introduces a novel intentionally biased bootstrap aggregating (biased bagging) method that prioritises informative data points by assigning them higher sampling probabilities during bootstrap sampling, in contrast to conventional bagging’s equal-probability approach. The proposed biased bagging method is designed and developed to increase the size of the small construction data sets artificially to achieve a better representation of the particular construction management prediction problem. Three base learners—artificial neural networks (ANN), support vector machines (SVM) and case-based reasoning (CBR)—are evaluated across ten construction datasets encompassing both regression and classification tasks. Comparative analysis indicates that biased bagging generally outperforms individual models, traditional bagging and the results reported in previous studies, yielding more accurate and reliable predictions.

Subject Keywords

Artificial neural network, Support vector machines, Case-based reasoning, Bagging, Ensemble learning

URI

https://hdl.handle.net/11511/115584

Collections

Graduate School of Natural and Applied Sciences, Thesis