Affective Computational Model to Extract Natural Affective States of Students with Asperger Syndrome (AS) in Computer-Based Learning Environment

This study was inspired by looking at the central role of emotion in the learning process, its impact on students’ performance; as well as the lack of affective computing models to detect and infer affective-cognitive states in real time for students with and without Asperger Syndrome (AS). This model overcomes gaps in other models that were designed for people with autism, which needed the use of sensors or physiological instrumentations to collect data. The model uses a webcam to capture students’ affective-cognitive states of confidence, uncertainty, engagement, anxiety, and boredom. These states have a dominant effect on the learning process. The model was trained and tested on a natural-spontaneous affective dataset for students with and without AS, which was collected for this purpose. The dataset was collected in an uncontrolled environment and included variations in culture, ethnicity, gender, facial and hairstyle, head movement, talking, glasses, illumination changes and background variation. The model structure used deep learning (DL) techniques like convolutional neural network (CNN) and long short-term memory (LSTM). DL is the-state-of-art tool that used to reduce data dimensionality and capturing non-linear complex features from simpler representations. The affective model provide reliable results with accuracy 90.06%. This model is the first model to detected affective states for adult students with AS without physiological or wearable instruments. For the first time, the occlusions in this model, like hand over face or head were considered an important indicator for affective states like boredom, anxiety, and uncertainty. These occlusions have been ignored in most other affective models. The essential information channels in this model are facial expressions, head movement, and eye gaze. The model can serve as an aided-technology for tutors to monitor and detect the behaviors of all students at the same time and help in predicting negative affective states during learning process.