Devising a ubiquitous solution for lie detection

Abstract

Lying, as always, remains a signi cant part of our day to day interactions covering both physical communication and digital communication using devices such as smartphones. However, to the best of our knowledge, an e ort is yet to be made to detect lying utilizing the ever increasing capabilities of smartphones. Therefore, in this paper, we investigate how far we can go in detecting lying through exploiting smartphones. To do so, rst, we judiciously develop a set of questionnaire that guarantees to indulge a person in providing a mix of true and false responses. Here, we develop a survey system worth of deploying in smartphones. The system, along with collecting the responses, accumulates corresponding usage data such as shaking, acceleration, tilt angle, etc. while holding the smartphone. Subsequently, after distinguishing false responses from true ones based on informal communication and other veri cations, we present distinguished responses and corresponding usage data collected from 47 participants to several machine learning algorithms. We nd that we can achieve from 72% to 81% accuracy in identifying false responses through analyzing the usage data using machine learning algorithms. Later, utilizing ndings of this analysis, we develop two di erent architectures for real-time lie detection using smartphones. Yet another user evaluation of the developed and implemented architectures con rms 84%-90% accuracy in lie detection.