Solving Hidden Non-markovian Models: How to Compute Conditional State Change Probabilities

Current production systems produce a wealth of information in the form of protocols, which is hardly exploited, often due to a lack of appropriate analysis methods. Hidden non-Markovian Models attempt to relieve this problem. They enable the analysis of not observable systems only based on recorded output. The paper implements and tests solution algorithms based on the original Hidden Markov Model algorithms. The problem of computing the conditional state change probabilities at specific points in time is solved. The implemented algorithms compute the probability of a given output sequence and the most probable generating path for Markov regenerative models. This enables the analysis of protocols of not observable systems and the solution of problems that cannot be solved using existing methods. This opens up new application areas, including a different approach to mining the wealth of data already collected today with huge monetary effort.