Improving Online Learning through a Deep Affective Computing with Predictive Analysis

Video recognition is challenging for Affective Computing(AC) because Data-Labeling tend to be noisy that reducing the accuracy of affective identifying. Furthermore, the delivery of just-in-time learning intervention via Predictive Analysis(PA) is still infancy. This study introduces a PA model based on the CNN-LSTM architecture, that applying a stack of layers with CNNs to read a sequence of images as input, LSTM for speech and text data-labeling in the middle, and MLP at the output to generate emotional classification. In addition, exploring Logistic Regression algorithm to forecast upcoming learning behavior and outcome with figure. We found a small improvement in learning productivity, however, the analysis surprisingly reveals our participants show high appreciation toward the intervention.