Classification of Coronavirus Images using Shrunken Features

Saban Ozturk; Umut Ozkaya; Mucahid Barstugan

The proposed framework includes three stages: feature extraction, over-sampling, and shrunken features. For this purpose, experiments are carried out in these three stages. Firstly, the classification results with 260 samples and 495 samples are examined for 78 features in raw form. The contribution of the over-sampling method is investigated by comparing the performance of the classification processes with the SVM algorithm. Table 1 shows the performance parameters of raw feature vectors consisting of 78 features of 260 samples and raw feature vectors of 495 samples created with the SMOTE algorithm. It is seen that increasing the minority classes with synthetic samples has a positive effect on classification accuracy. In Table 1 , it is seen that the classification made with the addition of synthetic classes creates a performance contribution of more than 10%. One of the most important reasons for this is the elimination of imbalance between . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.