Measuring Human Decision Confidence from EEG Signals in an Object Detection Task

In this paper, we investigate human decision confidence during image interpretation in an object detection task using electroencephalography (EEG) signals. We develop an EEG dataset acquired from 14 subjects. Five popular EEG features, differential entropy (DE), power spectral density (PSD), differential asymmetry (DASM), rational asymmetry (RASM) and asymmetry (ASM), and two classifiers, a support vector machine (SVM) and a deep neural network with shortcut connections (DNNS), are adopted to measure decision confidence in the object detection task. The classification results indicate that the DE feature with the DNNS model achieves the best accuracy of 47.36% and F1-score of 43.5% for five decision confidence levels. For the extreme confidence levels, the recognition accuracy reaches 83.98%, with an average Fl-score of 80.93%. We also found that the delta band performs better than the other four bands and that the prefrontal area and parietal area might be sensitive brain regions that represent decision confidence in object detection tasks.