TITLE:
A Sequential Logistic Regression-Gradient Boosting Framework for Diabetes Classification
AUTHORS:
Moyang Li, Bowen Cai
KEYWORDS:
Diabetes, Classification, Sequential Framework, Two-Stage Model, SHAP
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.9,
September
23,
2026
ABSTRACT: Diabetes mellitus, particularly type 2 diabetes, is a major global public health problem, making early detection and intervention essential for reducing disease incidence. This study used the Pima Indians Diabetes Dataset as the source of the training and internal validation sets and the Frankfurt Diabetes Dataset as the external test set. Since the outcome is imbalanced, with non-diabetic cases (Outcome = 0) as the majority class, a two-stage integrated model (a sequential two-stage classification framework) was proposed to reduce the false negative rate. Logistic regression first identifies high-confidence positive cases and immediately classifies them as diabetic; remaining patients are passed to GBM as a second stage. The final classification combines positive predictions from both stages, providing an additional opportunity to identify Outcome = 1 cases while reducing false-negative predictions. The two-stage model achieved an accuracy of 77.27% with a false negative rate of 38.89% on the internal validation set and an accuracy of 89.35% with a false negative rate of 20.47% on the external test set, demonstrating strong robustness and transferability. Additionally, SHAP was applied to enhance model interpretability, identifying Glucose and Insulin as the most important features, followed by BMI, DiabetesPedigreeFunction, and Age.