Modelling labour productivity using SVM and RF: a comparative study on classifiers performance

The purpose of this paper is to propose a data-driven approach for preparation of Construction Labour Productivity (CLP) models from influencing labour factors. Two state-of-art machine learning-based classifiers, Support Vector Machine (SVM) and Random Forest (RF) were used for modelling CLP. First, a preliminary review of previous literature was carried out to extract all CLP related factors. Subsequently, the list of CLP factors were ranked in terms of most influential in Malaysian Residential standpoint by experienced Project Managers through a pilot survey. The most influential factors identified were labour's lack of work experience, job category, education/training, nationality, skills, age and marital status. Data was collected based on these influencing factors from all construction workers in Malaysian Residential Projects. The data collected were used to develop CLP models using SVM and RF. The performance of the models was assessed using several statistical indices including Percentage of Correct (PC), Heidke Skill Score (HSS), the Probability of Detection (POD), the False Alarm Ratio (FAR) and the Peirce skill score (PSS). The SVM and RF simulated the CLP with high accuracy. The POD for both models was found above 90% in predicting different categories of productivity. The reliability plots showed a high efficiency of the models. The results indicate that the advanced machine learning methods can be used to achieve high accuracy in prediction of CLP. The present study can also be helpful for researchers and industry practitioners to understand how machine learning methods can be employed to learn more about productivity in construction and eventually improve the standards of construction labour productivity in Malaysia.