Adjusted Empirical Mode Decomposition with Improved performance for Signal Modeling and Prediction

An adjusted empirical mode decomposition method, built on Student's probability density function is presented. Compared to the original EMD, the new version provides a lower number of intrinsic mode functions and is more accurate in signal modeling and prediction. Using a backpropagation neural network for learning and in-sample prediction, our experimental results on a synthetic signal, an electrocardiogram (ECG), and a financial time series show that the presented tEMD is more efficient and leads to higher prediction accuracy than conventional EMD, regardless of the input time signal.