DEPENDENCE OF VITAMIN D LEVEL ON LABORATORY AND ANTHROPOMETRIC INDICATORS: APPLICATION OF MACHINE LEARNING METHODS FOR SCREENING IN ADULTS

Abstract

Vitamin D deficiency is now recognized as an international health issue, affecting a variety of physiological systems and disease outcomes. Purpose. The present study proposes machine learning models to identify individuals at risk of vitamin D deficiency. Materials and methods. Machine learning was used on the dataset of 944 persons’ laboratory analysis to determine the list of anthropometric and laboratory indicators that affect the development of vitamin D deficiency. It was built a decision tree with a depth of 5 to predict vitamin D deficiency based on various parameters. Results. The authors found feature importance in identifying potential vitamin D deficiency. Age and BMI were considered the most impactful anthropometric parameters, level of HDL was the most important laboratory parameter. A heatmap matrix for correlation of features between one another was created. It was calculated metrics based on the confusion matrix for determining the risk of a 25(OH)D deficit: Accuracy, Precision, Sensitivity, Specificity, F1-Score. The authors plotted the ROC curve of the optimal model; established that the Area Under the Curve (AUC) of the selected model is equal to 0.92 that is a very effective result. Conclusion. Machine learning techniques are more effective at predicting deficiencies than traditional statistical methods.