A Markov decision process with unobservable strength under Markovian environment

Altınkeser, Pınar
This thesis focuses on a system which survives or fails depending on the stress it is exposed to and strength occurrences. Stress and strength are random quantities generated by two factors, an unobservable Markovian environment and the actions applied. The distributions are known but only the resulting level of realized stress can be observed. The objective is to find an optimal policy that only uses available information to minimize the long run average cost of running this system. The base decision model is modified to handle lack of information, and a new model is built. Value of information is measured by comparing the optimal objective values of two models. Conflicting performance measures are also evaluated.