Dynamic duration estimate model using optimal weighted Least Square Moment Balanced Machine in preliminary design phase of construction

Purpose This study aims to develop a dynamic duration estimation model to accurately predict construction project duration during the preliminary design phase, where information is typically often relied on subjective judgment. Design/methodology/approach A hybrid machine learning model is proposed by integrating the Least Square Moment Balanced Machine (LSMBM) with the Artificial Satellite Search Algorithm (ASSA). Influential factors were identified through literature review, expert questionnaires and statistical correlation analysis, and the model was trained and validated on real project cases. Findings The proposed model achieved a mean absolute percentage error (MAPE) of 2.545%, indicating high predictive accuracy and outperformed existing basic and hybrid machine learning models. These findings verify the potential of optimization-driven AI to significantly improve the reliability of duration forecasting in early-stage construction planning to minimize time and cost deviations. Practical implications The proposed model offers construction managers a practical and data-driven tool for estimating project duration. By reducing dependency on subjective, experience-based estimation, the model supports more-accurate scheduling, better resource allocation and improved decision-making efficiency. Originality/value This study introduces a novel optimization-based LSMBM framework capable of producing accurate duration estimates in the early design phase. The integration of ASSA enables the automatic tuning of model parameters, reduces human bias and enhances model adaptability across different project conditions.