Continuous Mobile User Authentication Using a Hybrid CNN-Bi-LSTM Approach

作     者：Sarah Alzahrani Joud Alderaan Dalya Alatawi Bandar Alotaibi 

作者机构：Department of Information TechnologyUniversity of TabukTabuk71491Saudi Arabia Sensor Networks and Cellular Systems Research CenterUniversity of TabukTabuk71491Saudi Arabia 

出 版 物：《计算机、材料和连续体（英文）》 (Computers, Materials, & Continua)

年 卷 期：2023年第75卷第4期

页      面：651-667页

核心收录：

学科分类：0710[理学-生物学] 0810[工学-信息与通信工程] 12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by the Sensor Networks and Cellular Systems (SNCS)Research Center University of Tabuk Saudi Arabia under Grant 1443-001 

主　　题：Human activity recognition recurrent neural network(RNN) internet of things(IoT) machine learning(ML) 

摘      要：Internet of Things (IoT) devices incorporate a large amount ofdata in several fields, including those of medicine, business, and *** authentication is paramount in the IoT era to assure connecteddevices’ security. However, traditional authentication methods and conventionalbiometrics-based authentication approaches such as face recognition,fingerprints, and password are vulnerable to various attacks, including smudgeattacks, heat attacks, and shoulder surfing attacks. Behavioral biometrics isintroduced by the powerful sensing capabilities of IoT devices such as smartwearables and smartphones, enabling continuous authentication. ArtificialIntelligence (AI)-based approaches introduce a bright future in refining largeamounts of homogeneous biometric data to provide innovative user authenticationsolutions. This paper presents a new continuous passive authenticationapproach capable of learning the signatures of IoT users utilizing smartphonesensors such as a gyroscope, magnetometer, and accelerometer to recognizeusers by their physical activities. This approach integrates the convolutionalneural network (CNN) and recurrent neural network (RNN) models to learnsignatures of human activities from different users. A series of experiments areconducted using the MotionSense dataset to validate the effectiveness of theproposed method. Our technique offers a competitive verification accuracyequal to 98.4%.We compared the proposed method with several conventionalmachine learning and CNN models and found that our proposed modelachieves higher identification accuracy than the recently developed verificationsystems. The high accuracy achieved by the proposed method proves itseffectiveness in recognizing IoT users passively through their physical activitypatterns.