Machine learning-based dynamic model for on-site subcontractor performance management

Purpose – There is a necessity for a dynamic tracing, controlling, and process of decision-making for on-site subcontractor (SC) performance management during the project execution phase. Therefore, this study presents a dynamic model that offers a new way to SC management with the integration of machine learning (ML) for faster and more effective evaluation of on-site performance data of SCs. Design/methodology/approach – A literature review on both on-site SC performance evaluation and ML use in construction management practices was conducted. Then, in line with the gap in the literature, the model developing phase begins with the "On-Site SC performance measurement (PM)" and continues with the "subcontractor average weighted performance, " where criterion weights were considered through the Pythagorean fuzzy analytic hierarchy process and used in data entry for ML. The development of the model continues with "machine learning algorithm selection." The last stage consists of "the action plan" that constitutes the decision-making processes and is supported by expert support. Findings – For the ML-based model, six ML algorithms were tested individually, and decision tree algorithms were chosen among them and validated. The validation of the ML-based developed model was carried out on a superstructure project, and it was determined that the proposed model provided accurate results. The action plans suggested by the proposed model would help practitioners to determine corrective and/or precautionary actions in a faster and more accurate way regarding the real performance of SCs. Originality/value – This study lays stress on developing an ML-based dynamic performance management model based on the actual and continual PM of the SCs for the construction execution stage. Unlike existing literature that primarily focuses on selecting SCs based on their past performance during the bidding phase, this model enables real-time assessment of SC performance. In addition, with the help of ML integration, the dynamic structure of the model, which allows immediate identification of SCs who fall below the expected performance standards during the implementation phase and the derivation of relevant action plans, distinguishes the proposed model from other performance evaluation models.