Classification of Time Series Using FCM-based Forecasting Models

In this study, a method of time series classification is considered. Classification is performed using forecasting models. It is assumed that processed time series are of different natures, i.e., they belong to different classes. Each class has its forecasting model. Thus, an unknown time series is presented to the models to evaluate forecasting errors. The classified time series is assigned to the class with the winning forecasting model. In the study, Fuzzy Cognitive Maps are used to build forecasting models. Prior to forecasting, the processed raw time series are preprocessed. Six different error functions having the most significant influence on classification are used. The error functions come from root mean square error and mean percentage error.