Skip to Main Content
An APM classifier is considered for solving the time series classification task. Design, implementation and performance results are discussed. Two main components of the APM are the predictor and credit assignment modules. A system of predictive sigmoid neural networks trained offline is the key part of the predictor module. The credit assignment module is based upon an original credit update equation. No a priori statistical assumptions are required, and the neural predictors do not have to reach optimal training performance criterion to give good classification results. Although additional testing of the classifier is necessary to obtain final conclusions concerning its performance, the results obtained in this paper indicate that the APM classifier has the potential to solve certain classification problems in real time with a limited amount of training data.