Investigation of different ML approaches in classification of emotions induced by acute stress.

Background: Machine learning is becoming a common tool in monitoring emotion. However, methodological studies of the processing pipeline are scarce, especially ones using subjective appraisals as ground truth. New method: A novel protocol was used to induce cognitive load and physical discomfort, and emotional dimensions (arousal, valence, and dominance) were reported after each task. The performance of five common ML models with a versatile set of features (physiological features, task performance data, and personality trait) was compared in binary classification of subjectively assessed emotions. Results: The psychophysiological responses proved the protocol was successful in changing the mental state from baseline, also the cognitive and physical tasks were different. The optimization and performance of ML models used for emotion detection were evaluated. Additionally, methods to account for imbalanced classes were applied and shown to improve the classification performance. Comparison with existing methods: Classification of human emotional states often assumes the states are determined by the stimuli. However, individual appraisals vary. None of the past studies have classified subjective emotional dimensions with a set of features including biosignals, personality and behavior. Conclusion: Our data represent a typical setup in affective computing utilizing psychophysiological monitoring: N is low compared to number of features, inter-individual variability is high, and class imbalance cannot be avoided. Our observations are a) if possible, include features representing physiology, behavior and personality, b) use simple models and limited number of features to improve interpretability, c) address the possible imbalance, d) if the data size allows, use nested cross-validation.

Explainable stress type classification captures physiologically relevant responses in the Maastricht Acute Stress Test.

Introduction: Current stress detection methods concentrate on identification of stress and non-stress states despite the existence of various stress types. The present study performs a more specific, explainable stress classification, which could provide valuable information on the physiological stress reactions. Methods: Physiological responses were measured in the Maastricht Acute Stress Test (MAST), comprising alternating trials of cold pressor (inducing physiological stress and pain) and mental arithmetics (eliciting cognitive and social-evaluative stress). The responses in these subtasks were compared to each other and to the baseline through mixed model analysis. Subsequently, stress type detection was conducted with a comprehensive analysis of several machine learning components affecting classification. Finally, explainable artificial intelligence (XAI) methods were applied to analyze the influence of physiological features on model behavior. Results: Most of the investigated physiological reactions were specific to the stressors, and the subtasks could be distinguished from baseline with up to 86.5% balanced accuracy. The choice of the physiological signals to measure (up to 25%-point difference in balanced accuracy) and the selection of features (up to 7%-point difference) were the two key components in classification. Reflection of the XAI analysis to mixed model results and human physiology revealed that the stress detection model concentrated on physiological features relevant for the two stressors. Discussion: The findings confirm that multimodal machine learning classification can detect different types of stress reactions from baseline while focusing on physiologically sensible changes. Since the measured signals and feature selection affected classification performance the most, data analytic choices left limited input information uncompensated.

Eye movement behavior in a real-world virtual reality task reveals ADHD in children.

Eye movements and other rich data obtained in virtual reality (VR) environments resembling situations where symptoms are manifested could help in the objective detection of various symptoms in clinical conditions. In the present study, 37 children with attention deficit hyperactivity disorder and 36 typically developing controls (9-13 y.o) played a lifelike prospective memory game using head-mounted display with inbuilt 90 Hz eye tracker. Eye movement patterns had prominent group differences, but they were dispersed across the full performance time rather than associated with specific events or stimulus features. A support vector machine classifier trained on eye movement data showed excellent discrimination ability with 0.92 area under curve, which was significantly higher than for task performance measures or for eye movements obtained in a visual search task. We demonstrated that a naturalistic VR task combined with eye tracking allows accurate prediction of attention deficits, paving the way for precision diagnostics.

Sensory Appeal and Routines Beat Health Messages and Visibility Enhancements: Mixed-Methods Analysis of a Choice-Architecture Intervention in a Workplace Cafeteria.

Easier recognition and enhanced visibility of healthy options supposedly increase healthy choices, but real-world evidence remains scarce. Addressing this knowledge gap, we promoted nutritionally favourable foods in a workplace cafeteria with three choice-architectural strategies-priming posters, point-of-choice nutrition labels, and improved product placement-and assessed their effects on visual attention, food choices, and food consumption. Additionally, we developed a method for analysing real-world eye-tracking data. The study followed a pretest-posttest design whereby control and intervention condition lasted five days each. We monitored visual attention (i.e., total number and duration of fixations) and food choices with eye tracking, interviewed customers about perceived influences on food choices, and measured cafeteria-level food consumption (g). Individual-level data represents 22 control and 19 intervention participants recruited at the cafeteria entrance. Cafeteria-level data represents food consumption during the trial (556/589 meals sold). Results indicated that the posters and labels captured participants' visual attention (~13% of fixations on defined areas of interest before food choices), but the intervention had insignificant effects on visual attention to foods, on food choices, and on food consumption. Interviews revealed 17 perceived influences on food choices, the most common being sensory appeal, healthiness, and familiarity. To conclude, the intervention appeared capable of attracting visual attention, yet ineffective in increasing healthier eating. The developed method enabled a rigorous analysis of visual attention and food choices in a natural choice setting. We discuss ways to boost the impact of the intervention on behaviour, considering target groups' motives. The work contributes with a unique, mixed-methods approach and a real-world setting that enabled a multi-dimensional effects evaluation with high external validity.

Saccadic eye movements estimate prolonged time awake.

Prolonged time awake increases the need to sleep. Sleep pressure increases sleepiness, impairs human alertness and performance and increases the probability of human errors and accidents. Human performance and alertness during waking hours are influenced by homeostatic sleep drive and the circadian rhythm. Cognitive functions, especially attentional ones, are vulnerable to circadian rhythm and increasing sleep drive. A reliable, objective and practical metrics for estimating sleepiness could therefore be valuable. Our aim is to study whether saccades measured with electro-oculography (EOG) outside the laboratory could be used to estimate the overall time awake without sleep of a person. The number of executed saccades was measured in 11 participants during an 8-min saccade task. The saccades were recorded outside the laboratory (Naval Academy, Bergen) using EOG every sixth hour until 54 hr of time awake. Measurements were carried out on two occasions separated by 10 weeks. Five participants participated in both measurement weeks. The number of saccades decreased during sustained wakefulness. The data correlated with the three-process model of alertness; performance differed between participants but was stable within individual participants. A mathematically monotonous relation between performance in the saccade task and time awake was seen after removing the circadian rhythm component from measured eye movement data. The results imply that saccades measured with EOG can be used as a time-awake metric outside the laboratory.

Algorithm for automatic analysis of electro-oculographic data.

BACKGROUND: Large amounts of electro-oculographic (EOG) data, recorded during electroencephalographic (EEG) measurements, go underutilized. We present an automatic, auto-calibrating algorithm that allows efficient analysis of such data sets. METHODS: The auto-calibration is based on automatic threshold value estimation. Amplitude threshold values for saccades and blinks are determined based on features in the recorded signal. The performance of the developed algorithm was tested by analyzing 4854 saccades and 213 blinks recorded in two different conditions: a task where the eye movements were controlled (saccade task) and a task with free viewing (multitask). The results were compared with results from a video-oculography (VOG) device and manually scored blinks. RESULTS: The algorithm achieved 93% detection sensitivity for blinks with 4% false positive rate. The detection sensitivity for horizontal saccades was between 98% and 100%, and for oblique saccades between 95% and 100%. The classification sensitivity for horizontal and large oblique saccades (10 deg) was larger than 89%, and for vertical saccades larger than 82%. The duration and peak velocities of the detected horizontal saccades were similar to those in the literature. In the multitask measurement the detection sensitivity for saccades was 97% with a 6% false positive rate. CONCLUSION: The developed algorithm enables reliable analysis of EOG data recorded both during EEG and as a separate metrics.

Breakfast high in whey protein or carbohydrates improves coping with workload in healthy subjects.

Dietary components may affect brain function and influence behaviour by inducing the synthesis of neurotransmitters. The aim of the present study was to examine the influence of consumption of a whey protein-containing breakfast drink v. a carbohydrate drink v. control on subjective and physiological responses to mental workload in simulated work. In a randomised cross-over design, ten healthy subjects (seven women, median age 26 years, median BMI 23 kg/m(2)) participated in a single-blinded, placebo-controlled study. The subjects performed demanding work-like tasks after having a breakfast drink high in protein (HP) or high in carbohydrate (HC) or a control drink on separate sessions. Subjective states were assessed using the NASA Task Load Index (NASA-TLX), the Karolinska sleepiness scale (KSS) and the modified Profile of Mood States. Heart rate was recorded during task performance. The ratio of plasma tryptophan (Trp) to the sum of the other large neutral amino acids (LNAA) and salivary cortisol were also analysed. The plasma Trp:LNAA ratio was 30 % higher after the test drinks HP (median 0·13 (µmol/l)/(µmol/l)) and HC (median 0·13 (µmol/l)/(µmol/l)) than after the control drink (median 0·10 (µmol/l)/(µmol/l)). The increase in heart rate was smaller after the HP (median 2·7 beats/min) and HC (median 1·9 beats/min) drinks when compared with the control drink (median 7·2 beats/min) during task performance. Subjective sleepiness was reduced more after the HC drink (median KSS - 1·5) than after the control drink (median KSS - 0·5). There were no significant differences between the breakfast types in the NASA-TLX index, cortisol levels or task performance. We conclude that a breakfast drink high in whey protein or carbohydrates may improve coping with mental tasks in healthy subjects.