Back to Research
Dataset

Dataset Explorer

The model is trained on the CDC Behavioral Risk Factor Surveillance System (BRFSS) dataset — an annual telephone survey collecting health-related data across the US.

CDC BRFSS Survey

Source

14 clinical variables

Features

Binary diabetes diagnosis

Target

Feature Reference

Raw NameLabelTypeDescription
_BMI5BMIContinuousBody mass index (weight/height²)
_AGE80AgeOrdinalAge category 1–13 (18–24 to 80+)
SEXVARSexBinaryBiological sex (1=Male, 2=Female)
_IMPRACERace/EthnicityNominalRace/ethnicity category 1–6
GENHLTHGeneral HealthOrdinalSelf-rated health 1 (Excellent) to 5 (Poor)
PHYSHLTHPhysical Health DaysContinuousDays of poor physical health in past 30
SMOKE100SmokingBinarySmoked 100+ cigarettes in lifetime (1=Yes, 2=No)
_TOTINDAPhysical ActivityBinaryAny physical activity in past 30 days (1=Yes, 2=No)
EDUCAEducation LevelOrdinalHighest education level 1–6
INCOME3Income LevelOrdinalAnnual household income 1–11
_RFHYPE6HypertensionBinaryTold have high blood pressure (1=No, 2=Yes)
_RFCHOL3High CholesterolBinaryTold cholesterol is high (1=No, 2=Yes)
CHCKDNY2Kidney DiseaseBinaryTold have kidney disease (1=No, 2=Yes)
_MICHDHeart DiseaseBinaryHeart disease or heart attack (0=No, 1=Yes)

Ethics & Limitations

  • Data is self-reported — subject to recall and social desirability bias.
  • BRFSS over-represents certain demographics; model may not generalise equally across all populations.
  • Diabetes diagnosis in the dataset is self-reported, not laboratory-confirmed.
  • Socioeconomic features (income, education) are included as legitimate risk factors but require careful interpretation in SHAP analysis.