No One Acts like You: AI based Behavioral Biometric Identification

Abstract

There has already been a significant increase in Mobile Broadband (MBB) subscribers over the past few years, which is expected to increase further as a result of the current trend towards the Sixth Generation (6G) wireless systems. This is accompanied by a growing need for confidentiality, integrity and undoubtedly authentication of the communication units involved (human and machine). In particular, authentication is challenging with respect to human participants, as often only possession of a token or knowledge of a code/password is requested, which validates the underlying claim: who is the person. Furthermore, as knowledge-based passwords and object-based keys can be lost or stolen, biometric-based approaches rely on traits, which are intrinsically connected to one person. Therefore this work demonstrates the feasibility of Long Short-Term Memories (LSTMs) - a type of Neural Network (NN)-, to identify and authenticate people by their characteristic arm movements during specific everyday tasks. For measurements of the arm movement a wrist-band wearable has been developed. The focus of the evaluation with Deep Neural Networks (DNN) is on different high level structures of NNs and data. Multiple NNs are allocated to either different test subjects and user-activities. In a first task 1 out of n people had to be identified with a maximum average precision of 92.0 %. In a second evaluation the Networks performed in a n out of m authentication with an maximum average precision of 81.4 %. With those results the potential of NNs for biometric authentication with behavioral sensors is demonstrated. Additionally in view of the necessary personal biometric data, ethical and legal aspects are highlighted.