Brain–computer interface to predict impulse buying behavior using functional near-infrared spectroscopy

Figures

• 

  Figure. 1. Schematic channel location and measurement principle of fNIRS device: (a) indicates the location of fNIRS channels by attaching them to the participants’ forehead. Based on this channel location, the regions of interest (ROI) related to impulsiveness were set as shown in (b). (c) Shows the measurement principle by fNIRS. We can obtain optical density signals within banana-shaped pathways from detectors. The measured optical density signals are converted into the concentration changes of oxygenated (∆HbO) and deoxygenated (∆HbR) hemoglobin through the modified Beer-Lambert law.

• 

  Figure 2. Schematic illustration of the experiment protocol. We instructed the participants to perform Session 1 (the task of shopping at online duty-free shops) and Session 2 (the task of shopping at online shops). After the experiment, a self-reported measurement was conducted.

• 

  Figure 3. Average concentration of HbO for each channel. The blue bars are Session 1 and red bars refer to Session 2. Error bars indicate standard error of the mean (s.e.m) to estimate the variability of the mean concentration of HbO per channel. Stars indicate a statistically significant differences in mean concentration of HbO between Session 1 and Session 2 conditions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(^{***}<0.001; ^{**}p<0.01)$$\end{document}(∗∗∗<0.001;∗∗p<0.01). In all channels, hemodynamic differences between the two sessions are markedly different. This suggests the possibility of judging people's impulse buying tendencies.

• 

  Figure 4. Hemodynamic responses for the averaged change concentration of ∆HbO between Session 1 and Session 2. The ∆HbO represents the average interval from the start to the end of the task (0–60 s) acquired from 30 participants. The solid red line and dotted blue line are ∆HbOs of Session 1 and Session 2, respectively. The corresponding shaded colors are the standard deviations (s.d.) from the mean signals. The hemodynamic signals between the two sessions are markedly different (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t=3.297; ^{***}p<0.001$$\end{document}t=3.297;∗∗∗p<0.001), indicating the role of biomarkers for those who show impulse buying patterns.

• 

  Figure. 5. Classification accuracies of individual subject and grand averaged classification accuracy between Session 1 (impulse buying) and Session 2 (non-impulse buying) using SVM. All participants show the accuracies in the range of 83.91–99.49%. These results suggest that fNIRS data can be used as a biomarker to differentiate between impulse and non-impulse buying behaviors in terms of the BCI technology.

• 

  Figure. 6. Topographical maps of averaged ∆HbO activation at the PFC for 30 people: (a) shows a configuration of ROIs based on the Montreal Neurological Institute coordinates and their associated channels with the arrangement of 15 channels covering the forehead. (b,c) are averaged ∆HbO corresponding to Session 1 and Session 2, respectively. The color bars refer to the quantified brain activation within the range − 1 (low) to 1 (high). Session 1 shows the neural activity in mPFC, which is insignificant but active in the left VLPFC (Chs. 12, 14, 15). However, Session 2 illustrates a high activation in a wide range over the DLPFC, mPFC, and OFC except for the VLPFC region. The activation of mPFC and left VLPFC stems from the reward-related mechanisms, but the other areas are known to be responsible for inhibiting impulsiveness. This phenomenon is also statistically expressed, and a distinct ∆HbO difference has also been found between mPFC and the other areas (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt}…