Karthik Konda
Health analytics · Model comparison

Pima diabetes prediction

Compare three classifiers on a fictional record, then change the cutoff to see how recall and accuracy move on the test set.

Example record

For glucose, pressure, skinfold, insulin, and BMI, a blank or zero uses the training median. All inputs stay in this browser.

Positive-class model score
—

Logistic regression—
Random forest—
Soft voting—

Test the decision threshold

50%
Test recall
—
Test accuracy
—

Methods and data

The historical dataset contains 768 records for women of Pima heritage aged 21 and over. The three classifiers use eight measurements. The training split contains 614 records, and the stratified test split contains 154. The imputer and logistic standardization are fitted only on the training split.

The voting classifier averages the two component scores. A lower label cutoff can recover more positive cases while flagging more negatives. The displayed recall and accuracy come from the held-out benchmark records. This is an educational model comparison using fictional profiles.

Dataset and missing-value definitions ↗