Application of Machine Learning and Data Science in Project Construction Scheduling

Abstract

Construction scheduling is central to project success, but it remains challenging due to uncertain activity durations, resource interactions, supply chain variability, and frequent design or scope changes. Recent years (2020–2025) have seen rapid advances in machine learning (ML) and data science (DS) methods—ranging from gradient-boosted trees and deep sequence models to computer-vision pipelines—that augment or replace classical schedule-engineering techniques (CPM/PERT) by predicting activity durations and delay risk, automating progress measurement, and enabling dynamic, data-driven rescheduling. This paper (1) synthesizes literature from 2020–2025 on ML/DS applications in construction scheduling, identifying major thematic strands and research gaps; (2) proposes an end-to-end methodology combining supervised duration/delay models with a bi-objective resource-constrained project scheduling problem (RCPSP) that integrates ML predictions and uncertainty buffers; (3) demonstrates the approach on a reproducible synthetic dataset (temporal train/test split) and shows model performance (regression and classification) and schedule impacts; and (4) discusses implications, interpretability, and research directions. On the synthetic test set, a gradient boosting regressor reduced the mean absolute error (MAE) by approximately 16.6% compared to a coarse PERT-like baseline. A gradient boosting classifier for delay risk achieved an ROC-AUC ≈of approximately 0.98 and an F1 ≈score of approximately 0.89. When ML-predicted durations (risk-aware) were used in an illustrative RCPSP network, the planned makespan increased relative to optimistic PERT estimates—illustrating a trade-off between realism and nominal makespan. The review highlights recurring challenges: lack of open cross-project benchmarks, limited closed-loop rescheduling demonstrations, domain-shift and transfer issues, and a need for explainability and human-in-the-loop interfaces.