Does an LS-SVM classifier using EMD-derived features from RR-interval signals accurately classify diabetic versus normal patients?
A proposed machine learning methodology using RR-interval signals achieved 95.63% accuracy in classifying diabetic versus normal patients, offering a potential tool for early diabetes screening via ECG.
Diabetes Mellitus (DM) which is a chronic disease and difficult to cure. If diabetes is not treated in a timely manner, it may cause serious complications. For timely treatment, an early detection of the disease is of great interest. Diabetes can be detected by analyzing the RR-interval signals. This work presents a methodology for classification of diabetic and normal RR-interval signals. Firstly, empirical mode decomposition (EMD) method is applied to decompose the RR-interval signals in to intrinsic mode functions (IMFs). Then five parameters namely, area of analytic signal representation (AASR), mean frequency computed using Fourier-Bessel series expansion (MFFB), area of ellipse evaluated from second-order difference plot (ASODP), bandwidth due to frequency modulation (BFM) and bandwidth due to amplitude modulation (BAM) are extracted from IMFs obtained from RR-interval signals. Statistically significant features are fed to least square-support vector machine (LS-SVM) classifier. The three kernels namely, Radial Basis Function (RBF), Morlet wavelet, and Mexican hat wavelet kernels have been studied to obtain the suitable kernel function for the classification of diabetic and normal RR-interval signals. In this work, we have obtained the highest classification accuracy of 95.63%, using Morlet wavelet kernel function with 10-fold cross-validation. The classification system proposed in this work can help the clinicians to diagnose diabetes using electrocardiogram (ECG) signals.
Pachori et al. (Mon,) studied this question.