Sequential estimation and crashing in PERT networks with statistical dependence

We review a framework for learning-based estimation and dynamic decision in PERT networks. Our approach allows for probabilistic dependence on activity durations and revision of activity crashing decisions, whereas the traditional approaches like Program Evaluation and Review Technique make one-time crashing decisions based on the assumption of probabilistic independence on activity durations. Simulation results show that our dependent estimation and dynamic decision-making approach produces lower average project costs than the traditional independent estimation and static decision-making approaches. The effectiveness of our model is examined using sensitivity analysis, in that the expected relative benefit of our model over the traditional model tends to increase as the degree of correlation between activity durations increases.