Dynamic feature selection for accurately predicting construction productivity using symbiotic organisms search-optimized least square support vector machine

Productivity is one of the crucial elements for managing construction operations effectively which directly impacts on general cost and time of a project. The accurate prediction of productivity is thus of paramount importance to help construction manager give decision-making timely for avoiding cost overrun and project falling behind schedule. This study introduces an artificial intelligence (AI)-based inference model to accurately forecast productivity of a construction project. The model is a hybridization of least square support vector machine (LSSVM), symbiotic organisms search (SOS), and a feature selection (FS) method in which SOS proceeds the optimization process to achieve the greatest performance of LSSVM by simultaneously determining hyper-parameter of LSSVM model and set of highly relevant attributes of construction productivity. The performance of the proposed model, SOS-LSSVM-FS, is validated based on productivity dataset of two real projects located in Montreal, Canada constructed from September 2001 to June 2004. The statistical results of a 10-fold cross validation method indicate that SOS-LSSVM-FS achieves the highest accuracy of productivity prediction with 3.67% mean absolute percentage error (MAPE) which is at least 19.6% better than that of other comparative AI models. In addition, with the support of SOS, the model can run without human intervention of trial-and-error to tune control parameters. Therefore, this study contributes to core body of knowledge a novel model to deal with construction productivity-related problem. The SOS-LSSVM-FS model is strongly recommended as a promising tool for helping construction manager manage/control site productivity.

Automated summary

Productivity is crucial for managing construction operations effectively. The study introduces an AI-based inference model, SOS-LSSVM-FS, which achieves the highest accuracy of productivity prediction. With the support of SOS, the model can run without human intervention.