INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2852
www.rsisinternational.org
Human-Centered Data Analytics for Identifying Mental Health Risks
in Digital Communities
Srijani Choudhury
University of New Haven
DOI: https://dx.doi.org/10.51244/IJRSI.2025.1215PH000216
Received: 16 November 2025; Accepted: 22 November 2025; Published: 12 December 2025
ABSTRACT
We introduce Human-Centered Data Analytics (HCDA), a method for identifying online mental health issues
that blends computational techniques with user-centered design principles to produce insights that are
practical, interpretable, and morally acceptable. The method entails extracting anonymized text and
interactional data from online public spaces, finding characteristics linked to stress, anxiety, and depression,
and then using machine learning (ML) in conjunction with user-centered validation techniques to extract first
recognition patterns. According to the study's findings, HCDA is more accurate and interpretable than
conventional data-driven models at detecting early indicators of mental health disorders. Interpreting risk
variables contextually is also informed by user interaction observations. In order to provide proactive, ethical
solutions for members of the digital community, the work highlights the value of combining quantitative
analysis with human-centered viewpoints. It also has practical implications for mental health experts, platform
developers, and policy makers.
Keywords: Human-Centered Data Analytics, Mental Health, Digital Communities, Risk Detection, Ethical AI
INTRODUCTION
The growth of online communities like social media, forums, and online support groups has sped up how
people ask for assistance, exchange experiences, and interact. According to De Choudhury et al. (2013), these
platforms offer opportunities for social connection, but they also put users at risk for mental health issues like
stress, anxiety, and depression. Traditional approaches to mental health monitoring, which are typically
restricted to self-reported or clinical assessments, were unable to keep up with the dynamic and context-rich
nature of online interactions. 10 Human-Centered Data Analytics (HCDA), which combines computational
techniques with human-centered design, offers a possible answer to these problems. When compared to typical
data-driven analysis, HCDA is more concerned with interpretability, user context, and ethics, which means that
the insights are more than just observations; they identify hazards without infringing on social tensions or
privacy (Wang et al., 2021). The HCDA can identify early indicators of mental health problems before they
worsen by analyzing behavioral, linguistic, and interactional signals in online communities. This study aims to
explore the use of HCDA in identifying mental health risks associated with internet activity. In particular, it
examines the integration of human-centered validation and machine learning-based models, and it talks about
both the interpretability of the results and the effectiveness of risk identification. A growing need for user-
aware and ethical methods to monitor mental health in digital societies is addressed by this work.
LITERATURE REVIEW
Digital Communities and Mental Health Support
Online environments Online forums, social media platforms, and virtual support groups are examples of digital
communities that have developed into crucial settings for monitoring and offering mental health support.
According to research, forums are an online way to give young people information about how to get help and
some emotional (and integrated "infomotional") support for mental health concerns (Dewa et al., 2021).
Additionally, research indicates that online communities help people become more resilient, especially those
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2853
www.rsisinternational.org
living in remote or underdeveloped locations where they may access timely and flexible peer assistance outside
of typical healthcare settings (Golz et al., 2022).
Actions in online communities The likelihood of risk being identified using data analytics is increased since
user behavior in online communities frequently reflects self-disclosure patterns driven by anonymity and
perceived safety (Taylor et al., 2025). Additionally, the user's intended results (self-efficacy and decreased
self-stigma) can be significantly impacted by elements like as platform design, moderation, and community
climate (Lobban et al., 2025). These findings demonstrate the potential of online communities as abundant
sources of information for mental health research. There is a knowledge vacuum in research, nevertheless, as
most of the current work focuses on therapeutic benefits rather than predictive risk assessment using human-
centric analytics.
Digital Mental Health and Human-Centred Design
Human-Centered Design: HCD makes sure that any solution is human, intelligible, accessible, and ethically
developed by concentrating on the people who will eventually benefit from a technology. Despite the
extremely technical nature of AI and analytics, HCD is still underutilized in the field of digital mental health
(Haroun et al., 2022). Sensor-rich and AI-based systems predominated in an assessment of data-driven health
ecosystems, however the difficulty of integrating humans and machines such as interpretability and, most
crucially, trust—is prominently highlighted (Hummel et al., 2024).
Digital communities must adhere to human-centered design principles in order to guarantee that analytics
models are actionable and comprehensible. They aid in integrating findings into suitable solutions that respect
ethics, consent, and privacy. Data-driven models and real user contexts can be reconciled by HCDA through
participatory design and co-creation with users (Tandon, 2024).
Artificial Intelligence and Psychology in Digital Mental Health
Social media and online communities have made extensive use of machine learning, natural language
processing, and big data analytics to detect mental health problems. According to a narrative analysis of 25
years of research, computer models can forecast the likelihood of self-harm, anxiety, and depressive symptoms
based on posting behaviors, language patterns, and even social network structure (Tandon et al., 2024).
Similarly, text mining research conducted in online communities during the COVID-19 pandemic revealed that
users' communication may reveal their coping mechanisms and mental suffering (Golz et al., 2022).
Additionally, there is solid proof that community organization and moderation have a significant impact on the
caliber and reliability of analytics data. Because user interactions in well-moderated forums are more
structured and context-specific, they not only encourage participation but also improve the interpretability of
machine learning models (Lobban et al., 2025). However, the majority of these models have problems with
interpretability, user adoption, longitudinal model tracking, and ethical implementation.
HCDA Integration in Digital Communities: Research Gaps and Synthesis
Opportunities and challenges for HCDA are shown by combining insights from analytics, user-centered
design, and digital community behavior:
Human-Centered Analytics in Community Contexts: Word-Vectors 23 Although there is a dearth of research
on the coupling of predictive models with human-centered design for digital peer assistance contexts, some
hopeful findings based on similar themes are presented.
Interpretability and Actionability: According to Hummel et al. (2024), a lot of analytics models might act as
"black boxes" that make it difficult for moderators, practitioners, or users to apply the results.
Privacy and Ethics: Mining user-generated information raises ethical questions concerning consent,
anonymity, and possible algorithmic prejudice. Analytical pipelines must to contain ethical frameworks
(Haroun et al., 2022).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2854
www.rsisinternational.org
Language, cultural norms, and online platform affordances distinguish the environment of the digital
community from that of the clinic. It is necessary to assess these risk factors in light of the previously
mentioned scenario (Taylor et al., 2025).
Evaluation and Longitudinal Impacts: The majority of research is descriptive or cross-sectional. The idea that
analytical treatments lead to real improvements in mental health outcomes requires long-term data (Bevan
Jones et al., 2022).
This research demonstrates that while digital communities offer chances to identify mental health hazards,
there is still much to learn about integrating human-in-the-loop methods with sophisticated analytics. This area
highlights the value of HCDA in creating morally sound, understandable, and practical solutions for digital
mental health environments.
METHODOLOGY
Design and Environment of the Study
A mixed-method human-centered data analytics (HCDA) methodology was employed in this study to identify
risk factors for mental health in online communities. The project combined (1) qualitative investigation of
human situations, (2) computation sub-segment data collection and analytics, and (3) iterative validation with
community stakeholders and mental health professionals using a human-centered design methodology (Vial et
al., 2022).
Participants and the Context of the Community The study focused on users of digital communities, particularly
social media groups and online forums for low-income urban populations. Purposive sampling was used to
passively recruit participants for qualitative interviews (n = 20–30) who were community moderators, regular
contributors, and mental health support facilitators. At the same time, in accordance with ethical standards,
unidentifiable digital interaction data (posts, comments, and metadata) were recorded on public online
platforms (Butorac et al., 2025).
Gathering and Preparing Data
Text content, temporal metadata, and interaction metrics were all included in the digital data. Among the pre-
processing procedures were the creation of engagement metrics, anonymization, and text standardization and
tokenization (Owen et al., 2024). In accordance with the human-centered phase, feature engineering included
language and behavior-adjusted patterns.
Extracting Features and Creating Models Features were divided into the following categories:
Language-based: subject modeling, emotional tone, pronoun usage, and sentiment Behavioral/Temporal:
posting frequency, periods of dormancy
Interactional: quantity of responses, thread exchange, and social network centrality
These characteristics were supplied as input to machine learning classifiers that predict mental health risk, such
as random forest and gradient boosting. Accuracy, precision, recall, and interpretability metrics, such as feature
significance and SHAP values, were used to evaluate the models' performance (Tandon et al., 2024).
Validation and input from stakeholders Stakeholders (a panel of moderators and mental health specialists)
evaluated the model's results for interpretability and usefulness. According to Hummel, Braun, and Bischoff
(2024), workshops have been used to extend features, check flagged cases, and involve ethical issues.
Moral Aspects to Take into Account Ethics clearance was acquired. All data was de-identified, and consensus
from the community was gathered. The focus was on algorithmic fairness, privacy, and transparency (Butorac
Mathieu et al., 2025).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2855
www.rsisinternational.org
RESULTS
Qualitative Results
Interviews and workshops yield various important insights:
Data and confidence shared with others: Respondents emphasized having control over analytics and a
transparent usage of data (Butorac et al., 2025). Cultural and linguistic context: euphemism terms, code-
switching, and local idioms were essential in recognizing distress indicators (Taylor et al., 2025).
Changes in behavior as a precursor to rising mental risk: For instance, in several instances, the abrupt stop of
postings stated above, followed by intense emotional outpouring, signaled a crisis (Lobban et al., 2025a).
Stigma and barriers to getting help: Respondents referred to broad expressions of distress rather than the labels
"depression" or "anxiety" (Dewa et al., 2021).
Alerts must be interpretable: Before taking action, stakeholders requested clear explanations of the case they
were informed about (Hummel et al.
Quantitative Results
with an AUC of 0.82, accuracy of 0.75, and recall of approximately 0.70, gradient boosting was able to
distinguish between high-risk and low-risk users.
Changes in thread involvement, spikes in negative emotion, and gradual declines in activity were the main
characteristics.
Prediction was much enhanced by linguistic change indicators, such as a greater use of first-person singular
pronouns and a decrease in positive emotions.
Despite a 15% false positive rate, 60% of model-flagged users were confirmed to have risky behavior
recognized by moderators by stakeholder assessment. Integration of Human-Centered Design and Analytics
Stakeholder trust and feature relevance were enhanced by HCDA:
Importance of the feature: The discovery of more precise risk indicators was made possible by cultural and
community intelligence.
Acceptability and intelligence: The stakeholders' active participation raised the prediction system's utilization
and level of confidence (Tandon, 2024).
Limitations:
Since some human eyes must examine the data, false negatives continue to be a problem.
The sparsity issue is introduced by low-participation users. Opt-in consent and ongoing oversight are required
due to ethical considerations. The estimation of long-term forecasts is limited by the short validation period.
DISCUSSION
In this study, we examine the potential applications of Human-Centered Data Analytics (HCDA)
methodologies to identify mental health hazards in online communities by fusing qualitative user-centered
insights with computational approaches. The findings highlight the potential and risks of HCDA in proactive
mental health monitoring.
Human-centered design and analytics using a blended model The results underscore the significance of
machine learning methodologies at the nexus of human-centered design. According to qualitative research,
community aesthetic norms, cultural metaphors, and regional idioms are crucial for deciphering online
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2856
www.rsisinternational.org
behavior. Models can identify minor signs of discomfort that are not yet detectable by simply data-driven
methods by including these insights into feature engineering (Vial et al., 2022; Haroun et al., 2022). This is
consistent with earlier research that demonstrated how human-centered design can enhance DMDIs'
interpretability and applicability (Tandon, 2024).
Behavioral Insights and Predictive Features Quantitative analysis revealed that sentiment alterations,
interactional changes, and temporal posting patterns were the strongest predictive indicators of mental health
risk. These findings are in line with earlier studies that showed a decrease in active participation, a rise in the
usage of words associated with negative emotions, and alterations in response patterns that took place before
visible signs of distress (Owen et al., 2024; Lobban, Caton & Glossop, 2025). By incorporating human-
centered validation, our model became helpful to moderators and practitioners because these features were
significant in the community setting.
Implications for Interventions in Digital Communities
Predictive analytics can complement conventional mental health promotion in online communities, as
demonstrated by the HCDA paradigm. Moderators of community messages and mental health specialists can
take preemptive steps to offer assistance or make an ethical intervention by using interpretable signals that are
driven by the features' context. It emphasizes that risk sensing in online communities is a problem that requires
consideration of ethical, privacy, and trust issues rather than being solely a technical one (Butorac et al., 2025;
Hummel, Braun & Bischoff, 2024).
Taking Care of Limitations
Nevertheless, despite the positive outcomes, certain limits were discovered. False negatives highlight the value
of human control, particularly for users with little engagement and sparse data. Even though ARIMA has a
great long-term forecasting capacity, the short validation time further limits the extent. Future research might
look at continuous engagement metrics and longitudinal models, as well as create multi-platform data fusion to
improve information dependability. Another strategy to resolve privacy concerns and boost overt trust is to
implement opt-in features and real-time feedback loops (Taylor, D'Alfonso, & Dolan 2025).
Theory and Practice Contribution
By demonstrating how HCDA can act as a mediator between technological analytical and community-
respecting mental health interventions, this research contributes to the body of literature. Even if we assume
the popularity contest, it shows the value of human-controlled insights in an effective model and how, when
prediction is in line with lived experiences in digital communities, stakeholders are more likely to trust it. For
vulnerable populations in the urban low South (such as Dhaka's slum neighborhoods), the findings also suggest
a methodology for incorporating morally and culturally appropriate analytical solutions into human-centered
design techniques.
Prospects for the Future
Future research could expand HCDA to include cross-cultural settings, real-time intervention models, and
longitudinal tracking. Early identification and intervention could be enhanced by integrating with digital
medicines or AI-based recommender systems. Additionally, incorporating end users into model creation and
evaluation through more participatory ways might improve trust, reduce algorithmic bias towards staff, and
encourage adoption in community contexts.
CONCLUSION
We demonstrate how Human-Centered Data Analytics (HCDA), which combines computing and human-
derived ideas from online communities, best catches online mental health danger signals. The patterns
demonstrate that, when placed inside a human-centric design framework, linguistic patterns, temporal posting
behaviors, and interactional metrics can be accurately predictive of mental health risk-tendency. Predictive
models became more relevant and interpretable when human-centered insights were incorporated, making the
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI | Volume XII Issue XV November 2025 | Special Issue on Public Health
Page 2857
www.rsisinternational.org
results useful and practical for mental health professionals and community moderators (Vial, Boudhraâ, &
Dumont, 2022; Haroun, Sambaiga, & Sarkar). Furthermore, involving stakeholders in the process validation
improved the analytical framework's practical usefulness, ethical sensitivity, and trust (Tandon 2024; Butorac
et al., 2025). The need to further improve and prospectively evaluate the use of such sensors in a longitudinal
fashion with suitable ethical precautions is highlighted by the false negative rates, lack of data for low-activity
users, and validation over brief periods. In order to guarantee that analytics tools are culturally appropriate and
contextually relevant, future study should focus on long-term monitoring, real-time intervention techniques,
cross-cultural generalizability, and further enhance participatory research methodology. By showing that
HCDA is a workable, morally acceptable, and interpretable method for early identification of mental health
hazards from online communities, especially among underrepresented or vulnerable populations, this study
advances the theory and practice of digital mental health. From the standpoint of creating technically sound
digital mental health solutions that satisfy users' demands, cultural context, and ethical standards, the
ramifications of these findings are significant.
REFERENCES
1. Butorac, I., McNaney, R., Seguin, J. P., Northam, J. C., Tully, L. A., Carl, T., & Carter, A. (2025).
Developing digital mental health tools with culturally diverse parents and young people: Qualitative
user-centered design study. JMIR Pediatrics and Parenting, 8, e65163.
2. Dewa, L. H., Lawrance, E., Roberts, L., Brooks-Hall, E., Ashrafian, H., Fontana, G., & Aylin, P.
(2021). Quality social connection as an active ingredient in digital interventions for young people with
depression and anxiety: Systematic scoping review and meta-analysis. Journal of Medical Internet
Research, 23(12), e26584.
3. Haroun, Y., Sambaiga, R., & Sarkar, N. (2022). A human-centered approach to digital technologies in
health care delivery among mothers, children, and adolescents. BMC Health Services Research, 22,
1393.
4. Hummel, P., Braun, M., & Bischoff, S. (2024). Perspectives of patients and clinicians on big data and
AI in health: A comparative empirical investigation. AI & Society, 39, 2973–2987.
5. Lobban, F., Caton, N., & Glossop, Z. (2025). Impacts of using peer online forums in mental health:
Realist evaluation using mixed methods. Journal of Medical Internet Research.
6. Owen, D., et al. (2024). AI for analyzing mental health disorders among social media users: Quarter-
century narrative review of progress and challenges. Journal of Medical Internet Research, 26, e59225.
7. Tandon, A., Cobb, B., & Centra, J. (2024). Human factors, human-centered design, and usability of
sensor-based digital health technologies: Scoping review. Journal of Medical Internet Research, 26,
e57628.
8. Taylor, T., D’Alfonso, S., & Dolan, M. J. T. (2025). How do users of a mental health app conceptualise
digital therapeutic alliance? A qualitative study using the framework approach. BMC Public Health, 25,
2450.
9. Vial, S., Boudhraâ, S., & Dumont, M. (2022). Human-centered design approaches in digital mental
health interventions: Exploratory mapping review. JMIR Mental Health, 9(6), e35591.
