Front-End Cost Estimation by Selective Case-Based Reasoning for Building Construction Projects

Abstract

A successful building construction project can be achieved by estimating construction cost with high level of accuracy, which is particularly crucial in the front-end stage due to the influence on cost reduction and effective cost management. However, since there are problems regarding inaccurate budgeting for building construction projects, limited information availability, limited usage of unit price of actual construction cost, and lack of flexibility of cost estimation models for diverse building project, owners and cost estimators need to establish an effective cost estimation countermeasures.

To deal with the aforementioned problems, this dissertation aims to develop a front-end cost estimation methodology by selective case-based reasoning (CBR) for building construction projects for improving 1) cost estimation accuracy, 2) reliability of human trust on estimated costs, and 3) transparency of cost estimation process.

More specifically, this research has objectives of developing three modules that are 1) Module 1: Case-Base Development, 2) Module 2: CBR Method Selection, and 3) Module 3: CBR Cost Estimation. The Module 2 again comprises 1) Sub-Module 1: Normalization Method Selection (interval, Gaussian distribution-based, Z-score, logistic function-based, and ratio normalizations), 2) Sub-Module 2: Attribute Weighting Method Selection (Attribute Impact, entropy, feature counting, and genetic algorithms), and 3) Sub-Module 3: Similarity Measurement Method Selection (Mahalanobis distance-based, Euclidean distance-based, arithmetic summation-based, fractional function-based).

The proposed front-end cost estimation methodology by selective case-based reasoning was validated using leave-one-out cross validation method for multi-family housing (100 cases), military barrack (117 cases), and government office (52 cases) projects. Accuracy (under mean absolute error rate, mean squared deviation, and mean absolute deviation), stability (under standard deviation), and appropriateness (using kernel density estimation) of cost estimation results were examined. More importantly, the level of flexibility of the selective CBR model which provides the most accurate and stable normalization, attribute weighting, and similarity measurement method according to different types of building projects was tested.

The results of case studies for the validation of the proposed methodology are summarized as below: For the multi-family housing project, ratio normalization method, GA attribute weighting method, and arithmetic summation similarity measurement method-based CBR cost model was proposed to be the most accurate and stable. For the military barrack project, interval/ratio normalization method, GA attribute weighting method, and Euclidean distance similarity measurement method-based CBR cost model was suggested to be the most accurate and stable. For the government office project, ratio normalization method, AI attribute weighting method, and fractional function similarity measurement method-based CBR cost model was derived to be the most accurate and stable.

As contributions of the research, the suggested data preprocessed case-base development procedures are expected to improve transparency and reliability of cost estimate results. Also, this research performed validations of the improved estimate accuracy and explanatory power of the selective CBR models for different characteristics of case-bases. Consequently, accurate front-end cost estimations with enhanced flexibility responding to various building construction projects are expected.