Feature-Weighted Data Augmentation for Construction Cost Prediction

Accurate cost prediction in engineering projects is pivotal for informed decision-making since prediction accuracy determines resource allocation efficiency. Conventional regression analysis requires complete datasets; however, missing feature data - a pervasive issue in engineering practice - frequently leads to inadequate sample sizes. Current methodologies exhibit two inherent limitations: (1) ineffective capture of feature interaction effects, and (2) neglect of feature importance variability impacts on predictions. This study introduces a novel feature-weighting data augmentation (FWDA) framework. Feature importance scores are first quantified using random forest algorithms, followed by weight calculation. Weighted feature matrices are then generated through a strategic weighting process. An expanded feature space is constructed via polynomial feature synthesis, with subsequent orthogonal integration of original and synthesized features. Experimental results indicate a statistically significant reduction in prediction errors with corresponding accuracy improvements.