Towards Data Normalization Task for the Efficient Mining of Medical Data

The paper investigates the problem of data normalization in solving medical diagnostics tasks by machine learning algorithms. The authors describe five different data normalization methods' operations, advantages, and disadvantages. The effectiveness of their work was evaluated using two data sets with different Imbalanced Ratio, which is typical for medical tasks. The modeling was performed by solving a binary classification task using three different machine learning methods based on decision trees. It is experimentally established that the method of normalization ScalerOnCircle, unlike others, increases the efficiency of analyzing medical data based on researched machine learning methods. There was a significant increase in the F1-score value when using this normalization method. It is because ScalerOnCircle, in addition to normalization by columns, provides the possibility of considering relationships between the attributes of each vector of a given dataset. This problem is very acute in the medical field, where data sets designed for intellectual analysis are characterized by many attributes and complex nonlinear relationships between them. This fact must be taken into account when mining such datasets. ScalerOnCircle opens up several benefits for the efficient mining of medical data.