Symposium

Of Rising Scholars

Fall 2024

Ishan will be presenting at The Symposium of Rising Scholars on Saturday, September 21st! To attend the event and see Ishan's presentation.

Go to Polygence Scholars page
Ishan Mysore's cover illustration
Polygence Scholar2024
Ishan Mysore's profile

Ishan Mysore

Class of 2026S, CA

About

Projects

  • "Predicting Depression Risk for Adolescents through Social Media Post Data: A Machine Learning Approach" with mentor Kevin (Working project)

Project Portfolio

Predicting Depression Risk for Adolescents through Social Media Post Data: A Machine Learning Approach

Started June 13, 2024

Abstract or project description

How can social media data analysis and predictive modeling assist in prodromal symptom detection for depression in adolescents?

Title: Identifying Mental Health Issues in Adolescents through Social Media Data: A Predictive Modeling Approach

Introduction

The pervasive use of social media among teenagers is linked to various mental health issues. My project goal is to develop a predictive model using social media data to identify early signs of mental health problems for teens. This model will be used for prodromal symptom detection -- it will go through a teenager's social media account (Instagram, Twitter, YouTube, etc.) and it will flag any posts or view history items that indicate a mental health illness. Then, the model will advise a teenager to seek mental help and a diagnosis from a psychiatrist. Early detection through social media analysis can lead to timely mental health interventions, improving overall well-being.

Literature Review

Review research on how social media usage affects mental health in adolescents. Explore The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition to compile an accurate, official guide for classifying symptoms of each mental health disorder. Evaluate existing models that analyze social media data for mental health predictions. For the tests I’ve done so far, the three types of models that yield the highest accuracy are Support Vector Classifiers, Random Forest Classifiers, and K-Nearest Neighbors (when n>10). Also, explore ethical challenges in using social media data for mental health predictions.

Methodology

Collect anonymized social media post data from platforms like Twitter, Instagram, Tiktok, Facebook, and YouTube (from people experiencing mental health conditions and perfectly normal people). Collect anonymized social media view data from platforms like Twitter, Instagram, Tiktok, Facebook, and YouTube (from people experiencing mental health conditions and perfectly normal people). Generate example social media posts from the Amazon Claude 3.5 model, where some posts are normal and others contain hints of mental health disorders. Include metadata such as post frequency, sentiment, and engagement metrics.

Data Preprocessing

Clean and preprocess the data to remove noise and irrelevant information. Use natural language processing (NLP) to analyze textual content for emotional tone and psychological indicators.

Model Development

Apply machine learning algorithms (e.g., neural networks, support vector machines, random forest classifiers, k-nearest neighbors) to develop the predictive model. Train the model on a dataset labeled with known mental health outcomes.

Model Validation

Validate the model using a separate dataset to assess accuracy, precision, recall, and F1-score. Perform cross-validation to ensure model reliability and generalizability.

Ethical Implications

Implement robust data protection protocols and ensure informed consent. Monitor and mitigate biases to ensure fair treatment of all demographic groups. Ensure transparency in how the model operates and the limitations of predictions.

Applications and Recommendations

Integration with Mental Health Services: Discuss how the predictive model can be integrated with mental health support systems such as Scout by Sutter Health. Recommendation System: Develop a system to suggest personalized interventions based on model predictions. Do not use the model to diagnose someone; instead, use the model to indicate that something appears to not be right and that the person should seek mental health help immediately. Educational Outreach: Create educational materials for parents and educators on recognizing and responding to social media-related mental health issues.

Project Goals

Bare Minimum:

Create a predictive model that goes through a teenager's social media account, flags any content that doesn't seem normal, and (if necessary) advises the teenager to seek mental help and a diagnosis immediately.

Middle Ground:

(1) Create a platform that actually connects teenagers experiencing prodromal symptoms with nearby mental health professionals (based on the teen's location). Collaborate with global mental health organizations to validate and refine the model across diverse populations. Implement robust feedback mechanisms to gather user input and experiences, and use this feedback to iteratively improve the platform. OR (2) Implement this predictive model with Sutter Health's Scout app.

Reach Goal:

The model would not only predict mental health issues but also proactively prevent their escalation (should the teenager not have access to a nearby hospital immediately). Use predictive analytics to identify risk factors and suggest preventive measures before symptoms manifest.

  • Create an AI-driven platform that promotes positive mental health practices. The platform could encourage activities like journaling, mindfulness, and regular physical activity, tailored to each individual's preferences and needs.
  • Extend monitoring capabilities beyond social media to include other digital footprints such as smartphone usage patterns, wearable device data (e.g., sleep patterns, physical activity), and academic performance metrics.
  • Analyze situational factors influencing mental health, such as significant life events, changes in social relationships, and environmental stressors. Ensure the model is adaptable to different cultural contexts, recognizing and respecting cultural variations in the expression and perception of mental health.

Conclusion

Discuss the potential impact of the predictive model on early mental health interventions through social media analysis. Emphasize the importance of ethical considerations in the development and deployment of such models.