TITLE:
Predicting Diabetes-Related Survey Status Using Logistic Regression: Class Imbalance and the Effect of Classification Thresholds
AUTHORS:
Mizeng Tang
KEYWORDS:
Logistic Regression, Diabetes, Class Imbalance, Precision and Recall, Classification Threshold
JOURNAL NAME:
Journal of Data Analysis and Information Processing,
Vol.14 No.4,
November
18,
2026
ABSTRACT: The aim of this study was to evaluate how well logistic regression classifies diabetes-related status from health indicators in public survey-derived data. Using the UCI CDC Diabetes Health Indicators dataset, we compared a model using age category and BMI with an expanded model using selected demographic and health indicators. The share of records in the positive category, which combines reported prediabetes and diabetes, is 13.93%. The study also examined a class-weighted model and tested the impact of changing the classification threshold for predicting the outcome. The records were divided into three parts for stratified training, validation, and test purposes in proportions of 60%, 20%, and 20%, respectively. On the independent test sample, the expanded unweighted model had a ROC-AUC of 0.7966 and an average precision of 0.3640. Using a threshold of 0.50 resulted in a recall of 0.0941. However, selecting a threshold of 0.10, which was chosen on validation data to meet an illustrative 80% recall target, increased the recall to 0.8427 but reduced precision to 0.2530 along with an increase in false positives. Accuracy alone is not adequate for evaluating imbalanced classification problems, and an appropriate threshold should be set for the model’s purpose. This exploratory, unweighted secondary analysis is not a nationally representative estimate or a validated clinical screening tool.