Machine learning applications on laboratory test results

Vitamin D is among the essential vitamins for both adult and child health, and its deficiency is seen as a precursor to many diseases. Diabetes is a chronic health condition that plays a leading role in the formation of many fatal diseases. In this study, a diagnosis of vitamin D deficiency is proposed based on data mining techniques combined with feature selection. The study aims to accurately estimate the vitamin D value through blood test results, while also aiming to reduce the necessary parameters for estimation. Predictions made with machine learning give results incomparably faster than the vitamin D measurement process made in the laboratory. The best results were obtained as a result of the experiments completed on the data set of diabetes patients and it was seen that the highest classification accuracy was obtained with 98.030 % for the support vector machine model consisting of 19 features selected with the relif-f feature selection algorithm. As a result of the study, it has been seen that the number of the tests can be reduced. The performance of the methods was evaluated using performance metrics such as classification accuracy, sensitivity, specificity, precision, f-measure, and kappa. In addition, with this study, a novel benchmark data set was prepared for all parametric and non-parametric algorithms.