Development of a Classifier for the Diagnosis of Oncological Diseases Based on Blood Protein Markers

Abstract

The article deals with the problem of classifying oncological diseases by blood protein markers. To solve this problem, it is proposed to develop 2 data classifiers based on the SVM algorithm and a fully connected neural network using a three-class dataset containing normal patterns and patterns corresponding to 2 oncological diseases. At the same time, it is proposed to introduce dropout layers into the classifier based on a neural network in order to combat overfitting and use a special method for forming packages used in training to level the problem of class imbalance identified in the analysis of the dataset. The results of experimental studies show the advantage of the neural network-based classifier compared to the SVM classifier, which consists in a higher quality of the classification of minority classes corresponding to 2 oncological diseases.