Classification and prediction of diabetes disease using machine learning paradigm

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health

Background:

• Diabetes affects over 422 million globally, with projections reaching 642 million by 2040.
• This chronic condition, marked by high blood sugar, leads to severe complications like heart attack and kidney failure.

Purpose of the Study:

• To develop a machine learning (ML) system for predicting diabetic patients.
• Identify significant risk factors contributing to diabetes development.

Main Methods:

• Employed logistic regression (LR) for risk factor identification (p-value, odds ratio).
• Utilized four classifiers: Naïve Bayes (NB), Decision Tree (DT), Adaboost (AB), and Random Forest (RF).
• Applied K2, K5, and K10 partition protocols over 20 iterations, evaluating performance with accuracy (ACC) and area under the curve (AUC).

Main Results:

• Analyzed a dataset of 6561 respondents (2009-2012 NHANES), identifying 7 key risk factors including age, BMI, and cholesterol levels.
• The overall ML system achieved 90.62% accuracy.
• The LR feature selection combined with RF classifier yielded a 94.25% ACC and 0.95 AUC using the K10 protocol.

Conclusions:

• The combined logistic regression and random forest model demonstrates superior predictive performance.
• This ML approach offers a valuable tool for the early prediction of diabetes.