Analysis of Consumer Decision on E-Commerce Platforms with Video Advertising and Brain Data - Figure 9

Figures 9a and 9b depict the distribution of MAV_mean and SE_mean in the γ sub-band. The histograms clearly separate buy (blue) and not-buy (red) decisions with cutoff thresholds of 0.460 (MAV) and 1.735 (SE), Optimal cutoff thresholds were determined using Youden’s J index (Youden, 1950), which defines the ROC curve point maximising the sum of sensitivity and specificity (J = sensitivity + specificity – 1). The selected cutoff values achieved 97.1% classification accuracy with AUC ≈ 1.00 for the Gamma features, indicating near-perfect discrimination in this dataset. Given the modest sample size and limited stimulus set, this result should be interpreted cautiously and validated with larger, more heterogeneous samples to rule out overfitting. Figures 9c and 9d present β sub-band distributions. The MAV cutoff (0.460), obtained using Youden’s J criterion, achieved 85.0% accuracy (AUC = 0.866), while Shannon Entropy cutoff (1.095) achieved superior performance at 91.4% accuracy (AUC = 0.970), demonstrating that entropy provides stronger discriminative power than MAV in the β sub-band.