Although the high performance of the convolutional neural networks (CNNs) for brain-computer interface (BCI) tasks based on raw electroencephalography (EEG) signals, the explanation of the prediction result remains challenging owing to their complex structure and numerous parameters. We propose a novel framework for explaining CNNs for EEG-based BCI tasks by using the perturbation-based influence scores. The method supports the interpretation of CNN classification for EEG signals at both the example-level and the feature-level. The experiments on the BCIC III-IVa dataset demonstrate that the proposed method is effective for not only the interpretation of the predictive models, but also for the improvement of the classification accuracy.

There is a general lack of studies that are aimed at monitoring and detecting an operator's trust calibration, even though detecting someone's adjusted trust towards automation is essential to prevent misuse and disuse of automation. The goal of this article is to propose a convolutional neural network (CNN) based framework to estimate operators’ trust levels and detect their trust calibration in automation using image features of electroencephalogram (EEG) signals preserving temporal, spectral, and spatial information. Thirteen participants performed a set of automated Air Force multiattribute task battery tasks that differed in reliability (High/Low) and credibility (High/Low) levels. The proposed framework was compared with three machine learning methods—naïve bayes, support vector machine, multilayer perceptron—in terms of accuracy, sensitivity, and specificity of trust estimation and detection of trust calibration. Results of this article showed that the proposed framework had the highest performance of both trust estimation and detection of trust calibration in automation compared to the other comparison methods. This indicates that the proposed framework using the CNN classifier with the image-based EEG features could be an applicable model for estimating multilevel trust and detecting trust calibration during human-automation interaction. Also, it can help to prevent disuse and misuse of automation by estimating operators’ trust levels and monitoring their trust calibration in automation.