A Behavioral Biometrics-based Trust Framework for Continuous Crowdsensing Recruitment

Abstract

The inevitable trend of the large spread of smartphones and smart devices around the world has opened the door for a new sensing paradigm known as Mobile Crowdsensing (MCS). By recruiting workers with mobile devices to perform tasks over a cloud based MCS platform, organizations can collect sensing data in less time and at lower cost. One of the challenges faced in MCS systems which could prevent the task requester from obtaining reliable information is the presence of malicious workers who join the sensing task by impersonating other workers' identities. Identifying whether the data reports were submitted by a genuine user or an impersonator can help MCS systems exclude distrusted workers from the sensing task and therefore improve the quality and integrity of the submitted sensing reports. However, since MCS systems collect data from users while ensuring that their privacy is protected, the submitted reports cannot be linked to the workers which makes detecting impersonators in the system challenging. Behavioral biometrics refers to the unique behavioral traits that can be used to authenticate users based on how they naturally perform a specific activity. This work proposes a touch screen input behavioral biometrics-based trust framework that can support a reliable recruitment process in continuous MCS tasks. Using the touch screen input data collected from users in previous sensing tasks, unique machine learning models are built for each MCS worker which are then used to detect impersonators in the group during future sensing tasks. The proposed approach integrates the trained machine learning models with a dynamic continuous recruitment system taking into consideration the uncertainties accompanied with using the machine learning models. Simulations show that the proposed system improves the quality of the submitted reports when compared to a benchmark that only relies on users' historical performance.

Date of Award	Dec 2021
Original language	American English