Towards an Automated Deception Detection Based on Neurofeedback and Computational Techniques

Deception detection has been a controversial topic when analyzing human behavior, which many scientists call it as one of the most natural things among human beings. However in criminal interrogations, lie detection plays a major role in verifying evidence. Many of the previous researches had addressed this aspect in different approaches using various measurements such as heart rate, pupil size and facial expressions. Along with the developments in neurocomputing, many researches had started using more advanced psychophysiological measures such as EEG and fMRI to detect sudden changes in human physiology. However, due to the noise added to those measures by contextual factors such as emotions and distractions, those methods still suffer from getting false positive errors.


In this study, the effect of those contextual factors towards the process of deception detection has been addressed. The main intention was to study the impact of emotions and distractions against the process of questioning in a lie detection test. Hence, we created an artificial scenario to mimic a crime, and then analyzed the psychophysiological responses of the test subject by getting EEG and GSR values from Emotiv EPOC Neurohaedset and a custom GSR module created using Arduino Uno, respectively. The analysis was done by converting the EEG waves into the frequency domain in order to obtain the features of alpha, beta and theta rhythms. We also conducted an anonymous survey to obtain the attitudes of people towards lying, with the parallel intention of creating a proper design for our experiment.

After analyzing the features we obtained, it concluded that the measurements recorded during questioning differs according to the mind state of the test subject. We averaged the rhythmic intensities over 10 seconds duration for each questions to derive the differences. The theta variation in two questioning trials we conducted after each type of video, showed a significant difference of around 84%. Hence, we can conclude that this process can be used along with the already available deception detection methods in order to reduce the false positive errors in the questioning.

For the full desertion, please click on this link.

Leave a Reply