A Hybrid Evolutionary System for Parameter Optimization and Lag Selection in Time Series Forecasting

The accuracy of time series forecasting can be increased by the employment of evolutionary systems. The improvement in the precision of such systems impact positively on the decision making process of many organizations. In this work we explore the decomposition of time series into linear and nonlinear patterns by the use of an autoregressive integrated moving average (ARIMA) method and a support vector machine (SVM) method using a particle swarm optimization algorithm to simultaneously perform parameter selection and lag selection. Experiments were performed using datasets from the time series data library amd the results demonstrated that the proposed method achieved promising results for one-step ahead predictions.