Accurate and rapid screening model for potential diabetes mellitus

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning

Background:

• Early diabetes prediction is crucial for high-risk populations.
• Non-invasive clinical features can be utilized for diabetes risk assessment.

Purpose of the Study:

• To evaluate the effectiveness of five popular classifiers in identifying individuals with diabetes.
• To determine the best algorithm for diabetes classification using clinical features.

Main Methods:

• Utilized Weka data mining software to analyze 4205 annual physical examination reports.
• Assessed J48, AdaboostM1, SMO, Bayes Net, and Naïve Bayes classifiers.
• Employed nine non-invasive clinical features: age, gender, BMI, hypertension, cardiovascular disease history, family history of diabetes, physical activity, work stress, and salty food preference.

Main Results:

• The J48 decision tree classifier demonstrated superior performance with an accuracy of 0.9503.
• Key predictive features identified were age, family history of diabetes, work stress, and BMI.
• The model achieved high precision (0.950), recall (0.950), F-measure (0.948), and AUC (0.964).

Conclusions:

• Decision tree analysis offers a non-invasive method for early diabetes risk screening.
• This approach is valuable for developing regions and can aid clinical practitioners in rapid patient assessment.
• Identifying key risk factors can facilitate targeted community interventions for diabetes prevention and reduce healthcare system burden.