Journal of Resource Management and Decision Engineering In Press 2026 09 01 AI-Based Dynamic Resource Allocation in Construction Projects: A Reinforcement Learning Approach to Time–Cost–Resource Optimization AI-Based Dynamic Resource Allocation in Construction Projects: A Reinforcement Learning Approach to Time–Cost–Resource Optimization 1 19 EN Journal Article 2026 05 05 <p>This study aimed to develop and evaluate an artificial intelligence–based dynamic resource-allocation framework using reinforcement learning to simultaneously optimize project duration, cost performance, and resource utilization in construction projects. A quantitative computational study was conducted using data from 84 medium- and large-scale construction projects in Tehran, Iran, together with validation input from 168 construction professionals. Project-level data included activity schedules, labor and equipment allocation, progress, costs, resource conflicts, idle time, and operational constraints. A Proximal Policy Optimization reinforcement learning model was developed to dynamically allocate resources at weekly decision intervals. The model state incorporated schedule, cost, activity, and resource indicators, while the reward function jointly represented time, cost, and resource-efficiency objectives. Projects were divided at the project level into training, validation, and independent test subsets. Model performance was compared with historical managerial allocation and a conventional rule-based scheduling strategy. Robustness, sensitivity, and expert-validation analyses were also conducted. PPO-based allocation significantly reduced project completion time compared with historical allocation (mean difference = -2.68 months, 95% CI [-3.26, -2.10], p &lt; .001, d = 1.02) and rule-based allocation (-1.13 months, 95% CI [-1.48, -0.78], p &lt; .001, d = 0.71). Schedule deviation was significantly reduced by 7.69 percentage points relative to historical allocation (p &lt; .001, d = 1.56) and by 3.25 percentage points relative to rule-based allocation (p &lt; .001, d = 0.98). Cost deviation decreased by 5.83 and 2.42 percentage points, respectively (both p &lt; .001). Labor utilization significantly increased by 12.94 percentage points versus historical allocation (p &lt; .001, d = 1.91) and 6.22 percentage points versus rule-based allocation (p &lt; .001, d = 1.28), while equipment utilization also improved significantly. Resource conflicts were significantly lower under PPO than both comparison strategies (both p &lt; .001). Reinforcement learning provided a feasible, robust, and statistically superior approach to dynamic construction resource allocation, enabling simultaneous improvements in schedule performance, cost control, and resource utilization.</p> https://www.journalrmde.com/index.php/jrmde/article/download/453/421