PaperlayerView exact revision on arXiv

Source-linked AI summary

AI-Driven Analysis of the Effects of Recreational Activities to Well-Being using Physiological Responses: A survey

Aditi Site, Tarmo Lipping

arXiv:2609.03692v1eess.SP

TL;DR

Assessing well-being through recreation raises questions about how well-being is measured and which activities may help. This paper reviews AI-driven physiological-response studies and finds that real-life studies mostly use subjective questionnaires, whereas laboratory studies more often use wearable objective data.

  • Problem

    The review addresses how to define and measure well-being, which recreational activities may improve it, and what sensor-measurable factors matter, amid concern over mental and physical health.

    Source: §I · pp. 1–2

  • Method

    The authors conduct a narrative review of AI-driven analysis of physiological responses, using a literature search focused on peer-reviewed English-language articles relevant to the review’s themes.

    Source: §III · p. 4

  • Results

    Prior studies mostly used questionnaire-based subjective measures in real-life settings and wearable-based objective data in labs; subjective data used quantitative analysis, while objective data used machine learning.

    Source: 2) In lab environments · p. 6

  • Takeaways & Limitations

    The review positions well-being assessment as relevant to research and clinical practice and states that addressing well-being can help enhance user engagement and quality of life.

    Source: §VII · p. 9

    The review notes that mental-state prediction models still have room for performance improvement using more sophisticated methods, including deep learning or transformers.

    Source: 2) Well being · p. 9

Abstract

from arXiv · show

Well-being is a broader concept ensuring psychological, social, physical, and cognitive health. Recreational activities are one of the major approaches to facilitate well-being. There are various recreational activities in each of the domains in which well-being can be measured. Although well-being is often assessed using subjective scales, physiological responses such as heart rate, heart rate variability, skin responses and brain activity data provide objective ways to estimate the level of well-being. Research has focused on using these measurable physiological signals to develop AI driven well-being analysis concepts. This study presents a review of previous studies that have considered analyzing well-being. This review summarizes studies on the basis of recreational activities, data collection scenarios, subjective and objective data, data analysis methods and evaluation criteria. Furthermore, this study also proposes a framework to enhance data collection, analysis and visualization setup for the assessment of well-being.

I. INTRODUCTION · II. Conceptual background

The paper frames recreational activities as a way to support well-being and examines how AI can use physiological and behavioral sensor data to assess it. Its conceptual background links well-being, recreation, and physiological responses.

I. INTRODUCTION

  • Well-being encompasses social, cognitive, physical, and mental health, and recreational activities may improve it, particularly amid age-related health and mental-health risks.Recreational activities are leisure pursuits for enjoyment, relaxation, fun, or fitness.Source: §I · p. 1
  • Wearable sensors can noninvasively track physiological signals and behaviors, including heart rate, heart-rate variability, skin responses, brain activity, sleep, movement, and social interaction.The passage lists EEG, ECG, GSR/EDA, PPG, respiration, accelerometers, and gyroscopes among relevant measures.Source: §I · p. 1
  • Machine-learning approaches, including SVM, bagging, boosting, and deep learning, have been applied to sensor data to derive outcomes related to physical and mental health.These applications have contributed to AI techniques intended to facilitate overall well-being.Source: §I · pp. 1–2
  • The study asks how to define and measure well-being, which recreational activities may improve it, and which well-being factors can be measured with sensor data.The questions connect well-being components, activity domains, wearable-sensor indicators, and methods for affect modeling.Source: §I · p. 2
  • Monitoring physiological signals during recreation can support behavioral analysis and mental-state modeling, and the study considers recommending activities based on that analysis.The proposed possibility is a recommendation system intended to enhance the well-being effect of recreational services.Source: §I · p. 2

II. Conceptual background

  • The conceptual background presents well-being as an evolving concept shaped by sensor technologies and machine learning, with recreational activities central to its analysis.Prior studies have examined physiological responses collected during recreation to assess overall well-being.Source: §II · p. 2
  • The background organizes its discussion around the relationships among well-being, recreational activities, and physiological responses.Figure 2 highlights these three concepts and their connections before the following subsections explain them.Source: §II · p. 2

A. Well-being · B. Recreational activities

The paper focuses on psychological, social, physical, and intellectual well-being measurable through physiological responses, and describes recreational activities that can promote these dimensions.

A. Well-being

  • The study considers four physiologically measurable dimensions—psychological, social, physical, and intellectual well-being—within a broader multidimensional concept of quality of life.The perceived wellness model includes six components: psychological, social, intellectual, emotional, spiritual, and physical.Source: A. Well-being · p. 2
  • Psychological well-being concerns emotional and mental state and coping, while social well-being concerns meaningful relationships and community integration.Together with physical and intellectual well-being, these components are used to evaluate overall well-being.Source: A. Well-being · p. 2
  • Physical well-being concerns bodily health and functioning, while intellectual or cognitive well-being involves mentally stimulating tasks, creative and critical thought, and problem-solving.The passage describes physical well-being as including daily tasks without limitation and overall fitness.Source: A. Well-being · p. 2

B. Recreational activities

  • Recreational activities are voluntary leisure pursuits for enjoyment, relaxation, and breaks from routine, undertaken individually or in groups.The paper presents activities across psychological, social, physical, and intellectual dimensions.Source: B. Recreational activities · p. 2
  • Arts, culture, theater, singing, painting, travel, and nature walks can support social bonding and stress reduction, promoting social and psychological well-being.Source: B. Recreational activities · p. 2
  • Outdoor exercise and sports support physical well-being, while extreme sports contribute to recreational flow experiences.Source: B. Recreational activities · p. 2
  • Puzzles, sudoku, memory games, reading, and playing instruments can stimulate cognitive recreation and promote intellectual or cognitive well-being.Source: B. Recreational activities · p. 2

C. Physiological responses

Well-being assessment combines subjective scales with sensor-derived physiological measures; machine-learning models use these measures to estimate mental states, while the framework highlights unresolved links between measures, states, and well-being.

C. Physiological responses

  • Subjective assessment uses scales including the Perceived Wellness Scale, PANAS, STAI, and Self-Assessment Manikins, whereas objective assessment uses sensor data.Source: C. Physiological responses · pp. 2–3
  • Sensors including EEG, ECG, PPG, EDA/GSR, and accelerometers capture bodily responses from which heart rate, variability, skin conductance, movement, and brain activity can be derived.These variables can be used to assess or predict physical and mental state indicators.Source: C. Physiological responses · p. 3
  • Physical well-being is more straightforward to measure objectively; psychological, social, and intellectual well-being are commonly assessed by predicting mental states with machine-learning models –.Models are developed and trained using questionnaire-based subjective assessments of those mental states.Source: C. Physiological responses · p. 3
  • Open questions include whether subjective scales describe mental states well, how mental states relate to well-being, and how to address individual variability in well-being experience.The study aims to clarify these concepts and provide tools for exploring the questions more systematically.Source: C. Physiological responses · pp. 3–4
  • Well-being analysis requires an integrated setup combining recreational activities, wearable physiological sensors, data analysis and modeling, and subjective scales for model training and validation.Source: C. Physiological responses · p. 4

III. Methodology and organization

The paper uses a narrative review to examine AI-driven analysis of physiological responses in recreational activities and their effects on well-being. It organizes the review around data-collection contexts and types, analysis methods, and recommendation systems.

Methodology

  • The narrative review primarily searched IEEE Explore using combinations of well-being, recreational activities, physiological responses, machine learning, and behavior analysis, selecting relevant English-language peer-reviewed articles.It investigates physiological signals including EEG, ECG, GSR, and PPG.Source: §III · p. 4

Review organization

  • The review’s first two components cover real-life and laboratory data-collection scenarios, then physiological and subjective data and their collection methods.The review structure is shown in Figure 4.Source: §III · p. 4, Figure 4 · p. 4
  • The third component surveys qualitative, quantitative, and machine-learning analyses, including models, algorithms, and evaluation criteria for physiological responses.Source: §III · p. 4
  • The final component considers a comprehensive tool for recommending recreational activities and evaluating their effects on well-being; the literature analysis appears in Section IV.Source: §III · p. 4

IV. Comprehensive analysis of methods · A. Well-being and recreation

The review examines four components of well-being assessment to develop an AI-driven modeling methodology. It considers recreational activities and the settings, data, methods, and outcomes used in prior studies.

IV. Comprehensive analysis of methods

  • The review examines four key components of well-being assessment and aims to use prior studies’ outcomes and gaps to develop an AI-driven modeling methodology.The section is divided into four subsections, each analyzing one key point in well-being assessment.Source: §IV · p. 4

A. Well-being and recreation

  • Studies assess well-being in real-life recreational settings or in labs, where controlled stimuli and minimized external sensory input support more controlled data collection.Real-life examples include theater, concerts, and art exhibitions; lab activities include viewing media or completing cognitive, mental, or physical tests.Source: A. Well-being and recreation · p. 4
  • Table I organizes real-life studies by participants, activity, collection environment, data, sensors, analysis methods, and evaluations or outcomes.The passage describes the table’s organization, not comparative results.Source: 1) In real-life environments · p. 5, Table I · p. 5
  • Real-life studies involve diverse participants and activities reflecting psychological, social, physical, and cognitive aspects of recreation.Examples include younger and older adults, sports participants, retirement-community activities, extreme sports, and robot-assisted recreation.Source: 1) In real-life environments · p. 5
  • Studies use subjective questionnaires, including PIF, AFS, and PWS in, and participant experience self-reports collected through ESM in.In, the scales record socio-demographic characteristics, flow during recreation, and wellness level, respectively.Source: 1) In real-life environments · p. 5
  • Studies and also record motion and speech data to analyze participant interaction with other people and robots, respectively.The review therefore describes both questionnaire-based and behavioral data collection in real-life settings.Source: 1) In real-life environments · p. 5

1) In real-life environments: · 2) In lab environments:

Prior studies use different approaches across real-life and lab settings: real-life studies mostly rely on subjective methods, while lab studies more often use wearable-recorded objective data. Common physiological responses across both settings could support well-being modeling and emotion recognition in real life.

1) In real-life environments:

  • In real-life studies, measures used in include connectedness, SF-36, CES-D, and YPAS, covering overall well-being, depression screening, and physical activity.Source: 1) In real-life environments · p. 5
  • Real-life studies mostly use graphical, qualitative, or regression analysis; uses machine learning, while studies also test smartphones and robots for well-being assessment.Source: 1) In real-life environments · p. 5
  • In, researchers analyze the relationship between recreational flow experiences and perceived well-being during extreme sports.Source: 1) In real-life environments · p. 5

2) In lab environments:

  • Lab studies use public emotion-recognition datasets based on video clips,,, as well as stress, cognitive, and physical tasks; uses impromptu speech to study social anxiety in young adults.Table II summarizes lab studies assessing emotions and conditions such as anxiety or stress.Source: 2) In lab environments · p. 5, Table II · p. 6
  • Lab stress studies use settings or tests such as offices, automobiles, social stress, cold pressor, and Montreal imaging, recording mobility, PPG, EDA, heart rate, temperature, or EEG with wearables and sensors.Studies also use subjective scales for ground truth, and data analysis most commonly uses machine learning; examines anxiety in autistic children using walking tasks, and covers stress surveys.Source: 2) In lab environments · p. 6
  • Real-life studies mostly use questionnaire-based subjective methods, whereas lab studies more commonly use wearable-recorded objective data; both settings record movement, temperature, PPG, EEG, heart rate, and EDA.Subjective data are commonly analyzed quantitatively, while objective data are commonly analyzed with machine learning.Source: 2) In lab environments · p. 6
  • Common physiological responses across settings create potential for modeling well-being indicators and developing emotion-recognition models in real-life environments.Source: 2) In lab environments · p. 6

B. Data collection

This section reviews datasets used to model emotions, summarizing their participants, scenarios, devices, sensors, target variables, and subjective scales to inform data-collection protocols. It reports common collection patterns and annotation approaches for emotion assessment.

B. Data collection

  • The review covers WESAD, DREAMER, K-EmoCon, CASE, CLAS, and AMIGOS, describing cohort, scenario, device, sensor, target-variable, and scale information.The section notes that discusses video-based datasets used for emotion recognition in laboratory settings.Source: B. Data collection · p. 6, Table III · p. 7
  • Reviewed datasets averaged 30 participants aged 20–40; protocols most often used emotion-eliciting video clips, with some stress or cognitive tests.Common devices included Shimmer, Emotiv EPOC, and Empatica E4; objective measures included ECG, EEG, EDA or GSR, PPG, and accelerometer-based mobility, paired with subjective questionnaires such as PANAS and SAM.Source: B. Data collection · p. 7
  • Video-task annotations commonly used Russell’s circumplex model with valence and arousal axes, alongside emotion labels such as amusing, boring, happy, sad, and scary,.Source: B. Data collection · p. 7

C. Data analysis · D. Recommendation systems

The review contrasts quantitative analysis of subjective well-being data with machine-learning analysis of physiological and other objective data. It also surveys recommendation systems that suggest recreational activities to support well-being, especially social integration among older people.

C. Data analysis

  • Subjective data are analyzed mainly with quantitative methods, while objective data are analyzed using machine-learning methods.Quantitative approaches include statistical, regression, and graphical analysis; machine-learning methods model physiological responses to predict mental states.Source: C. Data analysis · p. 7
  • Studies, used statistical and regression analyses, deriving distribution metrics and relationships from limited questionnaire-based datasets.The reported metrics included skewness, kurtosis, mean, and standard deviation.Source: C. Data analysis · p. 8
  • Study used line plots and scatterplots to analyze speech frequency and conversation duration with robots.The review notes that this study, like the quantitative studies,, used limited questionnaire-based data.Source: C. Data analysis · p. 8
  • Studies,, used tree-based, boosting, regression, and SVM models to predict emotional states, social anxiety, or stress.The cited boosting models included XGBoost, light gradient boost, and CatBoost.Source: C. Data analysis · p. 8
  • Study combined k-means and fuzzy logic with real-time physiological signals to identify emotions at entrance gates.Other reviewed approaches included CNNs for speech-based mental-health assessment, Kalman filter methods, and customizable machine learning with feature selection.Source: C. Data analysis · p. 8

D. Recommendation systems

  • Recommendation systems can suggest activities based on behavior and well-being; study developed a physical- and social-care platform for older people.Its layers comprised sensing, data processing, knowledge gathering, and a smart assistant.Source: D. Recommendation systems · p. 8
  • Studies, recommended activities for older people using demographics, loneliness, interests, disabilities, and group size to support social engagement.Source: D. Recommendation systems · p. 8
  • Study proposed a context-aware mental-health recommender using sensors, user state and profile, and an intervention database, with a chatbot that learns from feedback.The review sees potential for end-to-end AI systems to assess well-being and recommend activities for people across age groups and needs.Source: D. Recommendation systems · p. 8

V. Future research direction · A. Research gaps · 2) Well being

The review identifies opportunities to refine well-being data collection, modeling, visualization, and recommendation, with future work targeting real-life recreational settings, multidimensional sensing, transparent visualizations, and stronger predictive models.

V. Future research direction

  • Prior well-being studies have used both real-life and laboratory environments, but data collection, modeling, visualization, and recommendation setups can be refined further.Source: §V · p. 8
  • The proposed framework aims to address gaps in data collection, modeling, visualization, and recommendation for well-being assessment.Source: §V · p. 8

A. Research gaps

  • Because most physiological measurement setups use laboratories, future protocols should record responses during recreational activities in natural or real-life environments.Source: A. Research gaps · p. 8

2) Well being

  • Well-being spans physical, psychological, social, cognitive, and mental perspectives, requiring sensors such as EDA, EEG, PPG, ECG, and respiration to relate variables to these aspects.Source: 2) Well being · p. 8
  • Interactive, transparent visualizations could improve system trust and adaptability by showing physiological changes in response to stimuli from real-life recreational activities.Source: 2) Well being · p. 9
  • Predictive models for anxiety, stress, emotions, and mood could be improved with more sophisticated methods, including deep learning or transformer models.Source: 2) Well being · p. 9

B. Execution framework

The proposed AI-driven framework combines recreational-activity recordings with physiological data and analysis outputs to support well-being assessment and visualization. Its tool is intended to help users recognize emotional patterns and stress-related states, support real-time applications, and assess audience reactions.

B. Execution framework

  • The interface aligns activity recordings with physiological time series, including raw sensor signals, derived measures, and predicted mental-state outputs.The advanced tool can display derived measures such as heart rate, heart rate variability, brain-wave entropy, and band power, alongside analyzed states such as anxiety, stress, and emotions.Source: B. Execution framework · p. 9
  • The tool helps users identify deviations from normal emotional patterns and recognize stress, anxiety, or arousal to enable self-regulatory behavior.Source: B. Execution framework · p. 9
  • Real-time outputs are described as essential for biofeedback and mental-health applications, while performers and artists can use the tool to examine audience reactions to scenes and events.Source: B. Execution framework · p. 9

VI. Discussion · VII. Conclusion

The review surveys how physiological responses can support AI-driven well-being analysis and proposes a framework for improved data collection, modeling, and visualization. It highlights the need for stronger evidence and scalable approaches, with potential relevance to clinical preventive care.

VI. Discussion

  • The proposed framework combines lab and real-life data collection using heart rate, skin conductance, heart rate variability, and activity data.It also proposes machine-learning models to recognize behavioral patterns, especially in real-life environments.Source: §VI · p. 9, Figure 5 · p. 10
  • The review covers collection environments, sensors, devices, analysis methods, evaluation criteria, and indicators including anxiety, depression, stress, emotions, and mood changes.It examines both laboratory and real-life data collection scenarios.Source: §VI · p. 9
  • The review identifies improved real-life data collection and machine-learning analysis as areas needing further research.The proposed protocol addresses this gap by combining multiple physiological responses with activity data across lab and real-life environments.Source: §VI · p. 9
  • A proposed real-time visualization framework would identify deviations in physiological responses to improve transparency and support mental health applications.The conclusion includes a simulation visualization tool.Source: §VI · p. 9, Figure 6 · p. 10
  • The findings may help clinicians in primary care and community settings identify well-being factors and implement early, low-intensity preventive interventions.The stated aim is to reduce progression to more severe mental health conditions.Source: §VI · p. 9

VII. Conclusion

  • The conclusion emphasizes well-being assessment as relevant to research and clinical practice, and as a means to enhance user engagement and quality of life.Source: §VII · p. 9
  • The review calls for continued research to strengthen the evidence base and develop effective, scalable AI-driven approaches to enhancing well-being.Source: §VII · pp. 9–10
Loading 2609.03692v1…