ECG Based Biometric Human Identification using Convolutional Neural Network in Smart Health Applications

In recent decade, the number of system uses biometric system for authentication. There were several biometric systems present based on external physiological characteristics such as face, iris, fingerprint, palm print etc. but few researchers worked on the internal physiological characteristics as a biometric. This project includes the design of an ECG-based biometric system that uses machine learning and deep learning techniques. ECG contains detailed information about electrical operation of the heart and the nature of this activity is highly personalized and can be used as biometric for authentication purpose. ECG based biometric can be mainly used in IOT based health care systems where data is transferred on internet. Other biometric systems require extra hardware to be used in health care systems. As ECG of a patient is taken in the hospitals, the same can be used for identification without extra hardware. Interval features of ECG signal are extracted and given to machine learning and deep learning algorithms. Machine learning techniques like SVM and KNN are used and deep learning is based on CNN. The datasets with diverse ECG behaviors are considered including MITDB, FANTASIA, NSRDB and QT. These datasets are collected from healthy or near-healthy participants and some include heart diseases such as arrhythmia and atrial fibrillation. The proposed CNN based approach achieved an accuracy of 81.33%, 96.95%, 94.73% and 92.85% on MITDB, FANTASIA, NSRDB and QT database respectively.