SVM based on personal identification system using Electrocardiograms

This paper presents a new algorithm for personal identification from their Electrocardiograms (ECG) which is based on morphological descriptors and Hermite Polynomials Expansion coefficients (HPEc). After preprocessing, we extracted ten morphological descriptors which were divided into homogeneous groups (amplitude, surface interval and slope) and we extracted sixty Hermite Polynomials Expansion coefficients (HPEc) from each heartbeat. For the classification, we employed a binary Support Vector Machines with Gaussian kernel and we adopted a particular strategy: we first classified groups of morphological descriptors separately then we combined them in one system. On the other hand, we classified the Hermite Polynomials Expansion coefficients apart and we associated them with all groups of morphological descriptors in a single system in order to improve overall performance. We tested our algorithm on 18 different healthy signals of the MIT_BIH database. The analysis of different groups separately showed that the best recognition performance is 96.45% for all morphological descriptors and the results of experiments showed that the proposed hybrid approach has led to an overall maximum of 98.97%. Keywords— Electrocardiogram (ECG), Hermite polynomials Expansion coefficients (HPEc), morphological descriptors, SVM, MIT_BIH database.