Modeling obesity using abductive networks

Area of Science:

• Medical Informatics
• Machine Learning
• Public Health

Background:

• Obesity, indicated by waist-to-hip ratio (WHR), is a significant health risk.
• Predicting clinical parameters from readily available data is crucial for efficient healthcare.
• Abductive networks offer a novel approach to medical data modeling.

Purpose of the Study:

• To investigate abductive network machine learning for modeling and predicting waist-to-hip ratio (WHR) using medical survey data.
• To assess the accuracy and efficiency of the AIM abductive network tool in modeling WHR.
• To explore the potential of this method for predicting other clinical parameters.

Main Methods:

• Utilized the AIM abductive network machine learning tool to model WHR from 13 health parameters.
• Trained models on 800 cases and evaluated on 300 cases from a Saudi Arabian primary healthcare sample (n=1100).
• Employed both continuous and categorical representations of parameters for model synthesis.

Main Results:

• Continuous WHR models predicted actual values within 7.5% error at 90% confidence limits.
• Categorical models achieved high accuracy, with only 2 errors in 300 cases.
• Analytical models explained population-level observations with up to 99% accuracy.
• Confirmed strong correlations between WHR and diastolic blood pressure, cholesterol, and family history of obesity.

Conclusions:

• Abductive networks provide a fast and automated method for medical data modeling, outperforming other statistical and neural network approaches.
• The developed models accurately predict WHR and offer insights into obesity-related health factors.
• This machine learning approach has broad applicability for predicting various clinical parameters in healthcare settings.