Swarm based Intelligent Feature Optimization Technique for ECG based Biometric Human Recognition

Recently, rapid growth in network technology and communication leads to widening human activities, so that a strong identification system is necessary. This work aims to build an accurate identification technology based on unique physiological characteristics of ECG. The Biometric recognition of ECG primarily depends on the quality of its features. Feature extraction is performed on parameters of a cardiac cycle based on the fiducial approach and large data sets of features have been extracted. The extracted dataset contains irrelevant, correlated and over-fitted features, which misleads the biometric system performance so that an effective feature optimization is needed to sort out those features to avoid redundancy in the data. In this paper, a novel swarm based intelligent feature optimization method; Particle Swarm Optimization (PSO) is used to generate feature subset based on a fitness function with joint entropy. The dataset from optimization phase are fed to classifiers such as ANN, K-NN and SVM for recognition. The proposed approach is tested with available open source MIT-BIH ECG ID database. Finally, a comparison is made with and without feature optimization in which PSO with KNN shows recognition accuracy of 97.8931%.