Time Series Classification and Forecasting Method Selection in the Task of Intellectual Systems Development

Nowadays, it is important to implement intellectual analysis methods in information systems of different areas. Processes in such systems are described by temporal data and can be presented in the form of time series. While analyzing those time series, one of the main tasks is forecasting. There are many forecasting methods and models known. This variety poses for a researcher a problem of choice of an efficient one. This paper proposes the model of the forecasting method selection based on valuable time series characteristics. The proposed model can be implemented as a multiclass classifier via neural networks such as a multilayer perceptron. Different multilayer perceptron architectures are presented in this paper. A comparison of their performances using the KFold method is also provided.