Abstract | ||
---|---|---|
A procedure for designing non-linear models for predicting time series is proposed. It is based on a set of rules emerging from a previously fitted ARIMA model. These rules are extracted from the set of coefficients in the ARIMA model, so they consider the autocorrelation structure of the time series, but a nonlinear approach is adopted. The proposed procedure is intended to help the user in the task of specifying as simple models as possible, providing an unambiguous methodology to construct machine learning models for time series forecasting. A generalization to time series with interventions is also proposed. The performance of these procedures is empirically studied by means of a comparative analysis involving time series from several domains and the multilayer perceptron is employed to approximate the non-linear models. |
Year | DOI | Venue |
---|---|---|
2005 | 10.1109/CIMCA.2005.1631617 | CIMCA/IAWTIC |
Keywords | Field | DocType |
autocorrelation structure,time series,nonlinear approach,simple model,non-linear modelling time series,arima fitting,proposed procedure,time series forecasting,arima model,multilayer perceptron,comparative analysis,non-linear model,machine learning,learning artificial intelligence,empirical study | Time series,Nonlinear system,Computer science,Autoregressive integrated moving average,Multilayer perceptron,Nonlinear modelling,Artificial intelligence,Machine learning,Autocorrelation | Conference |
ISBN | Citations | PageRank |
0-7695-2504-0-02 | 0 | 0.34 |
References | Authors | |
0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
R. Pino-Mejias | 1 | 0 | 0.34 |
E. L. Silva-Ramirez | 2 | 0 | 0.34 |
M. D. Cubiles-de-la-Vega | 3 | 0 | 0.34 |
M. Lopez-Coello | 4 | 0 | 0.34 |