Multimodal fusion; deep learning; affect recognition; explainability