Predicting the Onset of Diabetes with Machine Learning Methods

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning in Healthcare

Background:

• Diabetes prevalence is rising globally and in Taiwan, posing significant health and economic burdens.
• Diabetes complications can lead to severe disabilities and increased healthcare resource utilization.
• Early detection and prevention are crucial for managing diabetes and improving quality of life.

Purpose of the Study:

• To identify the most effective machine learning model for predicting diabetes risk.
• To evaluate the predictive performance of various parameters in diabetes diagnosis.
• To analyze outpatient examination data for diabetes risk factors in Taiwanese women.

Main Methods:

• Utilized a dataset of 15,000 women (aged 20-80) from a Taipei medical center (2018-2022).
• Investigated eight characteristics: pregnancies, plasma glucose, diastolic blood pressure, sebum thickness, insulin, BMI, diabetes pedigree function, and age.
• Trained and compared multiple machine learning models, including logistic regression, neural networks, decision jungle, and boosted decision tree, using Microsoft Machine Learning Studio.

Main Results:

• The two-class boosted decision tree model demonstrated superior predictive ability for diabetes.
• This model achieved an area under the curve (AUC) score of 0.991, outperforming other evaluated models.
• Key parameters like plasma glucose, BMI, and age were implicitly important in the predictive model.

Conclusions:

• The boosted decision tree model is highly effective for diabetes prediction in the studied population.
• Accurate diabetes prediction can aid in early intervention and resource allocation.
• Further research can explore integrating these models into clinical practice for proactive diabetes management.