Multimodal Psychological Dataset for Engineering Students in Art Therapy Research

In recent years, mental health issues among engineering students have become increasingly prominent, with academic stress and emotional suppression leading to widespread negative psychological states such as anxiety and depression. As a non-pharmacological intervention, art therapy is believed to have the potential to promote psychological recovery and enhance well-being, thanks to its emotional regulation and self-expression characteristics. However, there is currently a lack of multimodal psychological data on art therapy for engineering students, which limits in-depth research in this field. To address this, we present a multimodal psychological dataset for engineering students focused on art therapy research, systematically documenting the psychological changes and multi-source behavioral patterns of 26 participants before and after art therapy. The dataset follows a "pre-test—intervention—post-test" design, with questionnaires covering scales for anxiety, depression, subjective well-being, perceived stress, personality traits, and specific art therapy experiences. It also includes videos of the therapy process and subjective experience texts. Importantly, this dataset does not provide manually annotated emotion labels or behavior-category annotations derived from the video recordings.Instead, the video and audio data are released as raw, anonymized recordings intended to support downstream computational analysis, such as automated feature extraction, model-driven emotion recognition, and exploratory behavioral modeling by future researchers.Accordingly, no inter-rater annotation protocol or inter-rater reliability (IRR) metrics (e.g., Cohen’s Kappa or ICC) are reported in this study, as no human labeling process was conducted on the behavioral or emotional content of the recordings.This design choice was made deliberately to preserve the ecological validity of the intervention process and to avoid introducing subjective bias associated with predefined annotation schemas.The dataset is therefore positioned as a foundational multimodal resource rather than a fully annotated benchmark, allowing researchers to design and evaluate their own labeling strategies, annotation protocols, and reliability assessments according to specific research objectives. This dataset serves as a reusable empirical foundation for exploring the psychological mechanisms of art therapy, emotional expression patterns, and intervention effect prediction.