Rule-based algorithms with learning for sequential recognition problem

This paper deals with the sequential pattern recognition problem with dependencies among successive patterns, which undergo a control procedure. For this problem the original concept of recognition is presented in which two kinds of information are available: the learning set and the set of expert rules. Adopting the probabilistic model and assuming the first-order Markov dependence between patterns, the combined pattern recognition algorithm is derived. Additionally the concept of the unification algorithms, which transform the learning set into the rules and the expert rules into the learning set, are derived. The combined algorithm has been applied to the computer-aided diagnosis of human acid-base balance states and results of classification accuracies are given.