A Multimodal Data-Driven Framework for Anxiety Screening

Early screening for anxiety and the implementation of appropriate interventions are crucial in preventing self-harm and suicide among patients. While multimodal real-world data provides more objective evidence for anxiety screening, it also introduces redundant features that can lead to model overfitting. Furthermore, patients with anxiety disorders may not be accurately identified due to factors such as the fear of privacy breaches, inadequate medical resources in remote areas, and model interpretability, resulting in missed opportunities for intervention. However, the existing anxiety screening methods do not effectively address the outlined challenges. To tackle these issues, we propose an interpretable multimodal feature data-driven framework for noncontact anxiety detection. The framework incorporates an optimization objective in the form of a 0-1 integer programming function based on the ideal feature subset obtained from the feature selection component to enhance the model's generalization capability, which provides relevant diagnostic evidence of anxiety screening for psychiatrists. Additionally, a spatiotemporal feature reduction module is designed to capture both local and global information within time-series data, with a focus on key information within the time series to mitigate the influence of redundant features on anxiety screening. Experimental results on health data from over 200 seafarers demonstrate the superiority of the proposed framework when compared to other methods of comparison.