Skip to content

zhixin-lim/DSA4262

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 

Repository files navigation

DSA4262 — AI for Mental Health

Module: DSA4262 — Sense-Making Under Uncertainty | National University of Singapore
Theme: Applying data science to mental health — from population-level epidemiology to real-time NLP stress detection


Repository Structure

DSA4262/
├── Individual Assignment 1/
│   ├── data/
│   ├── DSA4262 Individual Assignment 1_Lim Zhi Xin.pdf   # Final report
│   ├── Individual Assignment 1_Lim Zhi Xin.Rmd           # R Markdown source
│   ├── plot1.png                                          # Gendered mental health burden
│   ├── plot2.png                                          # Gender paradox scatter
│   └── plot3.png                                          # Singapore age-stratified mortality
├── Individual Assignment 2/
│   ├── data/
│   │   ├── dreaddit-train.csv
│   │   └── dreaddit-test.csv
│   ├── DSA4262 Individual Assignment 2_Lim Zhi Xin.ipynb  # Full notebook
│   ├── DSA4262 Individual Assignment 2_Lim Zhi Xin.pdf    # Rendered report
│   ├── requirements.txt
│   └── README.md
└── README.md

Assignment 1 — Mental Health Disparities: When Prevalence Diverges from Mortality

Language: R (ggplot2, tidyverse)

Question

Do reported mental health prevalence rates translate proportionally into mortality risk — and what happens when they don't?

Approach

A three-stage visual narrative using Global Burden of Disease 2023 and WHO suicide mortality data, examining how gender disparities in reported prevalence map (or fail to map) onto fatal outcomes.

Plot 1 — Macro: Gendered Mental Health Burden
Grouped bar chart of disorder prevalence by gender. Women carry disproportionately higher rates of anxiety and depressive disorders; gender differences are minimal for bipolar disorder and schizophrenia. Establishes that prevalence-based narratives are shaped by help-seeking behaviour and detection bias, not underlying risk alone.

Plot 2 — Micro: The Gender Paradox
Quadrant scatter plot of male-to-female ratios for depression prevalence vs. suicide mortality across countries. Most countries cluster in the upper-left quadrant — men report lower depression prevalence yet experience substantially higher suicide mortality. The suicide axis spans a far wider range than the prevalence axis, revealing that modest diagnostic differences mask large disparities in fatal outcomes.

Plot 3 — Actionable: Singapore-Specific Prevention Gaps
Age-stratified suicide mortality for Singapore men vs. women, benchmarked against global trends. Two distinct high-risk patterns emerge among men: elevated mortality in young adulthood (despite high institutional visibility through schools and NS) and sustained elevation in later life (where institutional contact drops sharply). These represent two structurally different prevention failures.

Key Finding

Mental health prevalence, reported burden, and suicide mortality diverge across genders and life stages. The groups carrying the highest mortality burden are not the groups reporting the highest prevalence — meaning prevention strategies anchored to prevalence data alone will systematically under-serve the highest-risk populations.


Assignment 2 — Stress Detection in Social Media Text

See Individual Assignment 2/README.md for full details, setup instructions, and results.

Quick summary: Fine-tuned DistilBERT achieves F1 = 0.824 on the Dreaddit test set, outperforming structured LIWC features (0.756) and TF-IDF (0.740). Error analysis reveals the model under-detects stress expressed in neutral or relational terms — with implications for real-world deployment in mental health triage contexts.


Broader Theme

Both assignments engage the same underlying tension: quantitative models of mental health are only as useful as our understanding of where they break down.

Assignment 1 shows that aggregate prevalence data obscures the populations most at risk. Assignment 2 shows that NLP classifiers learn linguistic prototypes of stress rather than structural vulnerability. In both cases, the analytical contribution comes not from the performance metric alone, but from careful interpretation of what the data and model can and cannot tell us.

About

Stress detection in Reddit posts (DistilBERT, F1=0.82) + mental health disparity visualisation

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors