Machine learning regression for estimating the cost range of building projects

Purpose Several factors influence the costs of buildings. Thus, identifying the cost significant factors can assist to improve the accuracy of project cost forecasts during the planning phase. This paper aims to identify the cost significant parameters and explore the potential for improving the accuracy of cost forecasts for buildings using machine learning techniques and large data sets. Design/methodology/approach The Australian State of Victoria Building Authority data sets, which comprise various parameters such as cost of the buildings, materials used, gross floor areas (GFA) and type of buildings, have been used. Five different machine learning regression models, such as decision tree, linear regression, random forest, gradient boosting and k-nearest neighbor were used. Findings The findings of the study showed that among the chosen models, linear regression provided the worst outcome (r2 = 0.38) while decision tree (r2 = 0.66) and gradient boosting (r2 = 0.62) provided the best outcome. Among the analyzed features, the class of buildings explained about 34% of the variations, followed by GFA and walls, which both accounted for 26% of the variations. Originality/value The output of this research can provide important information regarding the factors that have major impacts on the costs of buildings in the Australian construction industry. The study revealed that the cost of buildings is highly influenced by their classes.