A new machine learning approach to select adaptive IMFs of EMD

Abstract

An adaptive algorithm for selection of Intrinsic Mode Functions (IMF) of Empirical Mode Decomposition (EMD) is a time demand in the field of signal processing. This paper presents a new model of an effective algorithm for the adaptive selection of IMFs for the EMD. Our proposed model suggests the decomposition of an input signal using EMD, and the resultant IMFs are classified into two categories the relevant noise free IMFs and the irrelevant noise dominant IMFs using a trained Support Vector Machine (SVM). The Pearson Correlation Coefficient (PCC) is used for the supervised training of SVM. Noise dominant IMFs are then de-noised using the Savitzky-Golay filter. The signal is reconstructed using both noise free and de-noised IMFs. Our proposed model makes the selection process of IMFs adaptive and it achieves high Signal to Noise Ratio (SNR) while the Percentage of RMS Difference (PRD) and Max Error values are low. Experimental result attained up to 41.79% SNR value, PRD and Max Error value reduced to 0.814% and 0.081%, respectively compared to other models.