Decision confidence estimation and electroencephalogram features analysis based on animal recognition task

Human-computer collaboration serves as a high-quality method to achieve optimal decisions in the workplace. However, there are relatively few existing papers that focus on how to effectively aggregate the viewpoints of different individuals. Contemporary research suggests that decision confidence bears a positive correlation with decision accuracy, thereby indicating that it can serve as a weighting parameter for aggregating viewpoints. This paper endeavors to quantitatively estimate decision confidence through the application of electroencephalogram (EEG). In this paper, we designed an animal recognition task to measure different decision confidence levels. The success of the task design was demonstrated by the comparison result of behavioral data and EEG at different confidence levels. In addition, then a neural network called channel attention based thinker-invariant DenseNet was proposed to predict confidence levels, with an average accuracy of 77.84%, higher than the results of existing models. Moreover, the regions of the brain associated with decision confidence, found by visualizing the channel attention module of our model, are consistent with existing studies.