Heartwave biometric authentication using machine learning algorithms

Abstract

The advancement of IoT, cloud services and technologies have prompted heighten IT access security. Many products and solutions have implemented biometric solution to address the security concern. Heartwave as biometric mode offers the potential due to the inability to falsify the signal and ease of signal acquisition from fingers. However the highly variated heartrate signal, due to heartrate has imposed much headwinds in the development of heartwave based biometric authentications. The thesis first review the state-of-the-arts in the domains of heartwave segmentation and feature extraction, and identifying discriminating features and classifications. In particular this thesis proposed a methodology of Discrete Wavelet Transformation integrated with heartrate dependent parameters to extract discriminating features reliably and accurately. In addition, statistical methodology using Gaussian Mixture Model-Hidden Markov Model integrated with user specific threshold and heartrate have been proposed and developed to provide classification of individual under varying heartrates. This investigation has led to the understanding that individual discriminating feature is a variable against heartrate. Similarly, the neural network based methodology leverages on ensemble-Deep Belief Network (DBN) with stacked DBN coded using Multiview Spectral Embedding has been explored and achieved good performance in classification. Importantly, the amount of data required for training is significantly reduced