Research / 2026

Research

I've worked on two independent research papers using public-health survey data, with guidance from Dr. Kiran Kumar Vejendla. Both began with a question about which factors matter most for health outcomes.

01 · Public health · May 2026

Income as a Stronger Predictor Than Education for Frequent Mental Distress in U.S. Adults: Evidence of Gender Disparities

Shubh Ghorpade · Curieux Review · 2026

I wanted to know whether income or education was more closely associated with how often adults reported poor mental-health days.

Data & approach

I analyzed 253,680 survey responses from a cleaned subset of the 2015 Behavioral Risk Factor Surveillance System (BRFSS). I used correlation, mean comparisons, and comparisons of frequent mental distress across groups.

What I found

Income had a stronger negative correlation with reported poor-mental-health days (r = −0.21) than education (r = −0.10). The relationships were modest; they do not establish cause and effect.

What this made me consider

Even if education is beneficial, it does not automatically remove the financial barriers that shape a person's living conditions or access to care.

Limits

The findings rely on self-reported, observational survey data and a cross-sectional analysis. They should not be interpreted as causal effects.

Read the full paper (PDF) ↗

02 · Machine learning · July 2026

Comparative Analysis of Mental Health & Physiological Predictors of Diabetes Using Machine Learning

Shubh Ghorpade · Curieux Review · 2026

I wanted to compare how well mental-health indicators and physical-health indicators predicted diabetes in the same survey dataset.

Data & approach

I used a balanced dataset of 70,692 records derived from the 2015 BRFSS survey. In Orange Data Mining, I compared logistic regression and random forest models with an 80/20 train-test split.

What I found

Logistic regression reached 74.6% classification accuracy (AUC 0.823). General health, high blood pressure, and difficulty walking were among the leading predictive features.

What this made me consider

A more complex model isn't always the better choice. I became more interested in model interpretation and what makes a predictor useful in practice.

Limits

This was a study of a historical, balanced public dataset—not a clinically validated screening tool. Performance should not be assumed to transfer directly to patient care.

Read the full paper (PDF) ↗