Mental Health Predictor

Background

• Defining Mental illness
  • Disturbance of thought , experience, and emotion that causes functional impairment in people.
  • Interpersonal difficulty, limiting the ability to work and self destructive behavior.
• In US nearly 1 in 5 adults aged 18 years or older (18.5%) have experienced mental illness (US Burden of Disease Collaborator, 2013).
• Data science can help us to better understand and effectively implement treatments for mental health problems
• Factors causing mental illness - anxiety, depression, biological, psychological and sociological (environment) approaches.

Project Focus

• Our analysis project focuses on identifying factors associated with the prevalence of poor mental health in the US.
  • Population density, income, water and land features
• Building a machine learning model that can predict mental health risk for an individual based on designated factors
• What are the most and least significant factors (features) in predicting prevalence of poor mental health in the US?

Description of Data Sources

• 500 Cities: Local Data for Better Health, 2019. 500 Cities: Mental health not good for >=14 days among adults aged >=18 years ---Centers for Disease Control and Prevention (CDC), Division of Population Health, Epidemiology and Surveillance Branch
  • Behavioral Risk Factor Surveillance System (BRFSS) data (2017, 2016)
    • Mental Health Severity: Respondents aged ≥18 years who report 14 or more days during the past 30 days during which their mental health was not good.
• US Household Income Statistics---Golden Oak Research Group LLC, “U.S. Income Database Kaggle”. Publication: 5, August 2017
• United States Cities Database---SimpleMaps.com, Pareto Software LLC, compiled data from U.S. Geological Survey and U.S. Census Bureau

Tools/Resources

• Creating ERD
  • QuickDBD
• Creating Database
  • SQLite
• Analyzing Data
  • Pandas
• Connecting to Database
• Machine Learning
  • Imbalanced-learn
  • Scikit-Learn
  • Tensorflow
  • Dashboard
  • Tableau
  • Flask
  • HTML/CSS

Machine Learning Model

• We used a binary outcome based on %poor mental health prevalence.
• The binary outcome was calculated by median split:
  • The median % poor mental health of the 500 cities was 13.89%. So…
    • If a city < 13.89% poor mental health → “Good Mental Health”
    • If a city >= 13.89% poor mental health → “Bad Mental Health”
• Features were log-transformed and scaled to bring them into a normal distribution
• We tried logistic regression, support vector machines (1-3 kernels), decision tree, gradient tree boost (learning rates .05 - 1), random forest, and 1-2 layer deep learning
• We used 10-fold cross-validation - i.e., 10 machine learning instances of randomly allocating 90% of data to training and 10% to testing. We averaged the performance across the 10 instances.

Summary of Findings

• Multiple machine learning models were used and most of them provided about 80% accuracy in their mental health risk prediction.
• With Random Forest model the most strongest feature in predicting poor mental health was Standard Deviation of Income.
• This suggests that income inequality in a city most predicted the prevalence of poor mental health.

Limitations

• Mental health data
  • Small sample
  • Limited availability
  • Subjective self-rating
• Differences in time frame of the datasets

Recommendations for Further Analysis

• Exploring data related to:
  • Impact of COVID-19
  • Climate
  • Affordable healthcare
  • Availability of mental healthcare providers
  • Data from wearable devices can be used to identify physiological markers associated with mental illness
  • Social media and internet activity can provide insight into a behavioral side of mental health

Dashboards

• Tableau Dashboard
  • Data Exporation Visuals and Machine Learning Summary