Improving robustness of case-based reasoning for early-stage construction cost estimation

In the long-term use of the Case-based reasoning (CBR) model for early-stage construction cost estimation, a typical issue is the unstable knowledge structure when the actual data distribution does not satisfy the assumed distribution. This study combines Modal Linear Regression (MODLR) with CBR, for comparison with the con-ventional CBR models using genetic algorithm (GA) and ordinary least squares (OLS), tested by simulated data and a case study of 1610 apartment buildings. The results show the variance of attribute weight in MODLR-CBR is far less than others, validating its superior knowledge stability in dealing with changes in the case-base. This study not only bridges the gap in the robustness of CBR models, but also prepares construction cost practitioners tackle the massive growth in the volume of the cost data. The results can be further referenced to the area of multidimensional optimization in CBR.