Joint Optimization of Learning and Project Abandonment Decisions

This study aims to address the significant financial losses and consequences that can result from unexpected system failures during project execution. The duration of each project is random with varying distribution parameters that cannot be directly observed. The primary focus is on determining optimal project abandonment policies for integrated parameter learning of random project duration. To identify the optimal abandonment policy and infer unknown parameters, a parametric Bayesian framework is employed. The problem is cast into a partially observable Markov decision process framework to minimize the expected costs associated with project failures and system failures. Through an analysis of the structural properties of the value function, we establish the existence of an optimal abandonment threshold, leading to a state-dependent control limit policy. Furthermore, we examine the existence and monotonicity of this control limit to balance cost reduction with optimal performance.