A combined Parametric and Nonparametric Approach to Time Series Analysis
The analysis and prediction of natural phenomena is an interesting and challenging task. Time series obtained from the observation of one or more features of a phenomenon are often the only access to the data generating system.
Unfortunately, time series analysis is usually done by specialists in the field of the phenomenon with traditional analysis techniques. The application of modern analysis and prediction tools is often avoided due to their complexity or the risk of failure. This issue can be surmounted by an interdisciplinary approach.
This work is an example for the possible synergetic effect of interdisciplinary research. In the field of oceanography the coastal upwelling phenomenon is analysed in experimental studies with a numerical model in order to develop a parametric prediction model. Artificial neural networks seem to be a suitable parametric model.
However, in the field of computer science traditional artificial neural techniques showed limitations in the analysis and prediction of time series obtained from natural phenomena, particularly with nonlinear and nonstationary time series. Motivated by this limitations a new approach to time series analysis and prediction is presented in this work, the mixture of nonparametric segmented experts (MONSE).
The MONSE approach is exploiting the synergetic effect of a combined nonparametric and parametric analysis. It is supposed to be applied to explorative time series analysis and prediction in various fields, i.e. in a context where hardly any kowledge about the time series of concern is available.