A hill-climbing approach to construct near-optimal decision trees

We consider the problem of identifying the state of an n component coherent system, where each component can be working or failed. It is costly to determine the states of the components. The goal is to find a decision tree which specifies the order of the components to be tested with minimum expected cost. The problem is known to be NP-hard. We present an extremely promising heuristic method for creating effective decision trees, and computational results show that the method obtains optimal solutions for 95% of the cases tested.<
>