Nowadays, considering society’s highly demanding lifestyles, it is important to consider the usefulness of relaxation from the perspective of both psychology and clinical practice. The response towards relaxation (RResp) is a mind-body interaction that relaxes the organism or compensates for the physiological effects caused by stress. This work aims to automatically detect the different mental states (relaxation, rest and stress) in which RResps may occur so that complete feedback about the quality of the relaxation can be given to the subject itself, the psychologist or the doctor. To this end, an experiment was conducted to induce both states of stress and relaxation in a sample of 20 university students (average age of $$25.76\pm 3.7$$25.76 ± 3.7 years old). The electrocardiographic and electrodermal activity signals collected from the participants produced a dataset with 1641 episodes or instances in which the previously mentioned mental states take place. This data was used to extract up to 50 features and train several supervised learning algorithms (rule-based, trees, probabilistic, ensemble classifiers, etc.) using and not using feature selection techniques. Besides, the authors synthesised the cardiac activity information into a single new feature and discretised it down to three levels. The experimentation revealed which features were most discriminating, reaching a classification average accuracy of up to $$94.01\pm 1.73$$94.01 ± 1.73 % with the 6 most relevant features for the own-collected dataset. Finally, being restrictive, the same solution/subspace was tested with a dataset referenced in the bibliography (WESAD) and scored an average accuracy of $$90.36\pm 1.62$$90.36 ± 1.62 %.

In-the-wild eye tracking datasets to train gaze vector estimation methods are noisy and unreliable, due both to limited image quality and to weak guarantees on label accuracy. At the same time, simply increasing dataset size does not guarantee better results, especially when diversity, balance and label reliability are not taken into account. In this work, we adopt a data-centric perspective and introduce an innovative data-pruning pipeline for robust gaze vector estimation. Our methodology leverages vision-language model embeddings and their zero-shot classification capabilities, together with an early-learning prediction-error analysis, to assign task-aware quality scores to individual samples. Moreover, we incorporate a region-weighted loss conditioned on gaze direction to better handle the uneven difficulty across the gaze space and to improve learning in under-represented regions. Finally, we validate our approach on a large relabeled subset of the GazeCapture dataset by training a hybrid gaze-estimation model, and show that a curated subset of 84K samples yields an accuracy improvement over a model trained on the full 2.45M-sample dataset (https://dx.doi.org/10.21227/c9ap-zm59 doi:10.21227/c9ap-zm59).

Monitoring autonomic nervous system responses offers significant potential for personalized health applications and early intervention. This study presents a machine learning framework for binary classification of physiological activation states using electrocardiogram (ECG) and respiratory (RESP) signals from the WESAD dataset. Stress and amusement were grouped as Activation, whereas baseline and meditation were grouped as Non-Activation. The processing pipeline incorporates R-peak detection, temporal segmentation, handcrafted feature extraction, and subject-independent evaluation. Conventional machine learning classifiers and deep learning architectures were evaluated, together with a proposed Hybrid Multi-Scale Temporal Convolutional–Conformer (MST-Conformer) combining temporal representations learned from ECG and RESP waveforms with handcrafted physiological features. Performance was assessed using 15-fold Leave-One-Subject-Out (LOSO) cross-validation. Adding respiration-derived features to ECG-derived features improved conventional machine learning performance, increasing the highest balanced accuracy from 0.752 to 0.807 and AUC from 0.802 to 0.886. The complete Hybrid MST-Conformer incorporating 27 physiological features achieved a mean accuracy of 0.825, balanced accuracy of 0.820, F1-score of 0.763, and AUC of 0.904. Subject-level analysis also revealed substantial physiological heterogeneity. These results support the potential of combining learned temporal representations with handcrafted physiological information for subject-independent physiological activation classification.