WARM-VR: a Wearable Affect Recognition from Multisensory stimuli in Virtual Reality Dataset

With the growing integration of human-computer interaction into everyday life, advances in machine learning have enabled systems to better perceive and respond to users\u2019 emotional states. Affective computing, in particular, has emerged as a key area of research, leveraging continuous physiological signals from wearable sensors to detect and measure human affect. However, most existing affect recognition datasets focus on static environments, limiting their applicability to immersive contexts such as virtual reality (VR). In this paper, we introduce WARM-VR (Wearable Affect Recognition from Multisensory stimuli in Virtual Reality), a novel publicly available dataset designed to support affect recognition in multisensory environments with wearable sensing instrumentation. Data were collected from 31 participants aged 19\u201337 using wearable sensors: a wristband measuring blood volume pulse (BVP), electrodermal activity (EDA), skin temperature (TEMP), and motion (ACC), and a chest strap recording electrocardiogram (ECG) signals. Participants engaged in VR experiences designed to elicit relaxation through a calming beach environment following stress induction via an arithmetic task. These sessions incorporated synchronized visual, auditory, and olfactory stimuli. Affective states were assessed subjectively through validated self-report questionnaires and objectively through the analysis of physiological measurements. Statistical analysis of the questionnaires confirmed that VR relaxation significantly reduced negative affect, particularly with olfactory enhancement. Furthermore, we established a benchmark on the dataset using widely recognized machine learning algorithms. The best performance for binary classification from BVP data of valence state was achieved using a CNN architecture (F1 score: 0.64, AUC: 0.72). For arousal, a CNN combined with bidirectional LSTM yielded the best results (F1 score: 0.62, AUC: 0.66). Notably, GRU-based models consistently outperformed others in the relaxation classification task.