Multimodal physiological signals from wearable sensors for affective computing: A systematic review

In recent years, advances in wearable devices have driven the rapid development of multimodal physiological signals for affective computing. In this paper, we systematically review 34 relevant studies between 2015 and 2025 following the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines, focusing on the combination of multimodal wearable sensing with learning models. We find that multimodal measurements have become increasingly common in the field, with electrodermal activity (EDA) and electrocardiograms (ECGs) being the most commonly used signal modalities. Several widely used public datasets, including WESAD, DAPPER, and DEAP, provide a basis for model optimization and comparative experiments. Deep learning methods have been widely adopted, with convolutional neural networks (CNNs) and long short-term memory (LSTM) being the most commonly used, and gradually extending to architectures such as the transformer. Despite recent advancements in multimodal emotion recognition studies, the field still faces key challenges, including the lack of long-term continuous data in naturalistic settings, the limited adaptability of current models to multimodalities, and the limited generalizability and interpretability of models for real-world applications. Future research needs to continue to address data acquisition, method optimization, and model generalizability in applications.