A Predictive Performance Analysis of Vitamin D Deficiency Using a Decision Tree model

Abstract

Background. HBV infection is a major health problem which may be life-threatening. Vitamin D (VD) is involved in various pathophysiological mechanisms in a plethora of diseases. And also, there is a strong demand for the prediction of its severity using different methods. The study aims to evaluate performance of DT as one of the machine learning models in the prediction of severity in vitamin D deficiency. Methods. In total, data containing serum VD levels were collected from 292 CHB patients. The independent characteristics such as: age, sex, weight, height, zinc, BMI, body fat, sunlight exposure, and milk consumption were used for prediction of VD deficiency. 60% of them were allocated to a training dataset randomly. To evaluate the performance of decision-tree the remaining 40% were used as the testing dataset. The validation of the model was evaluated by ROC curve. Results. The prevalence of VD deficiency was high among patients (63.0%). The final experimentation results showed that DT classifier achieves better accuracy of 96 % and outperforms well on training and testing of VD dataset. Also, the areas under the ROC curve AUC is 0.78, when we applied DT algorithm with the significant variables by cross validation, the values of AUC = 0.78 and 85.3% accuracy were obtained. Conclusion. We concluded that the serum level of Zn is an important associated risk factor for identifying cases with vitamin D deficiency. Also, the risk of VD deficiency could be predicted with high accuracy using decision tree learning algorithm that could be used for antiviral therapy in CHB patients.