Machine Learning-Driven Prediction of Vitamin D Deficiency Severity with Hybrid Optimization

Area of Science:

• Biomedical Informatics
• Machine Learning Applications
• Nutritional Science

Background:

• Vitamin D deficiency (VDD) poses global health risks.
• Standard 25-hydroxy vitamin D (25-OH-D) blood tests are often impractical.
• Non-invasive prediction of VDD severity is increasingly needed.

Purpose of the Study:

• Develop a clinically acceptable machine learning (ML) model for VDD detection.
• Eliminate the need for 25-OH-D blood tests.
• Address model overfitting and enhance multi-class prediction.

Main Methods:

• Applied data reduction, cleaning, and transformation to the vitamin D dataset.
• Utilized the improved whale optimization (IWOA) algorithm for feature selection and weight function optimization.
• Employed a stacking classifier for enhanced prediction accuracy.

Main Results:

• Achieved a high accuracy of 99.4% in VDD detection.
• The proposed IWOA algorithm demonstrated superior performance in feature selection.
• The stacking classifier proved effective and robust compared to other models.

Conclusions:

• The developed ML model offers a promising non-invasive approach for VDD assessment.
• Advanced optimization techniques significantly improve prediction accuracy.
• The stacking classifier is a robust choice for detecting vitamin D deficiencies.