Classification performance analysis of combined fNIRS-polygraph system using different temporal windows

Polygraph has been widely used to detect lie which measures some physiological parameters in human body but many researchers believe that lie can be detected more precisely if we can find the neuronal changes which occur during a deception process. To investigate the hemodynamic response related to deception, functional near-infrared spectroscopy (fNIRS) is used at the prefrontal cortex (PFC) of the subject. Commercially available polygraph system, Paragon acquisition system (PAS) is used to measure the physiological parameters like respiration and electrodermal activity (EDA) during the deception process. A mock crime scenario with a stimulus of single trial is used as a deception protocol. The acquired data is classified into true and lie classes, based on hemoglobin concentration changes in case of fNIRS and based on the physiological signal changes for polygraph data. Linear discriminant analysis (LDA) is used as a classifier. In this paper, we have analyzed the effect of different temporal windows in classification of combined fNIRS-polygraph signals for lie detection. The results show that the best time window for fNIRS-polygraph system is 2-7 s and 0-7 s for fNIRS and polygraph data respectively.