Ensemble Empirical Mode Decomposition for time series prediction in wireless sensor networks

This paper outlines the use of Ensemble Empirical Mode Decomposition (EEMD) as a preprocessing step in wireless sensor network time series prediction using support vector machines. Inherent adaptive data analysis approach of the decomposition process makes the system robust to signals driven from non-linear and non-stationary processes. We propose two variants of the hybrid model called EEMD-SVM and EEMD-SVM-SUM and compare them with the stand-alone use of support vector machines for one-step ahead prediction. Root mean square error and correlation coefficients are used for performance comparison. Results indicate that the hybrid models enhance prediction accuracy as the original complex sensed phenomenon is decomposed into several simpler components which reduces the computational complexity of the support vector machines and increases their class separability.