Main Article Content
The purpose of the research was to design and develop guidelines for using heartbeat signals in biometric applications. The heartbeat signals used in the research consist of two parts. The first part was taken from the MIT-BIH database. The second part, which included new heartbeat signals was collected by dividing the signals into cycle. Each cycle (1 signal period) comprised of P-wave, QRS complex and T-wave. Experiments were carried out in such a way that different referenced lines at different positions on the signals, namely 0.6, 0.7, 0.8 and 0.9, were used to find the position values that appeared suitable in each data set. The characteristics of the signal were then extracted with the discrete wavelet transform technique (DWT) The characteristics obtained would later be used for experimenting with forward feed-back-propagation artificial neural networks in order to find the correctness of each process. Another method used in the experiments was the reverse leaning (Back-Propagation) together with the Scaled Conjugate Gradient method for training the data. These experiments were carried out to find the number of neural nodes and the number of hidden layers that would be suitable for the heartbeat signals data.
The results showed that: 1) The two parts of the heartbeat signals were: the 10 signals from the MIT-BIH database with Record100, Record101, Record103, Record113, Record116, Record117, Record123, Record202, Record205 and Record209; and the signals collected from the Neulog Ecg Nul-218 usb of twenty users, whose average age were 24.8 years old. 2) The most suitable reference line was found to be at 0.7, because it gave the highest average accuracy of 96.97%. This also corresponded with the experiment done on the MIT-BIH dataset. Which gave the accuracy of 93.35%. 3) The results of the neural node number test showed that 200 neural nodes provided the highest accuracy. This would, therefore, be suitable for identification purposes. The signals from the MIT-BIH database hed the highest average of 96.68% and the newly collected heartbeat from that sample gave the highest accuracy rate of 95.29%