Analysis Accuracy of Complementary Label for Multi-Class Classification: Case Study on Chest X- Ray Images

Collecting label data for multi-class data is always costly and time-consuming. To overcome this problem, we put forward an effective method, namely complementary labels for multi-class classification. As opposed to the regular classification in which the original labels are given to training data, learning with complementary labels specifies one of the classes that the pattern does not belong to. Collecting ordinary labels will be more exhausting than collecting complementary labels because users must find out the ordinary class from some of the candidate classes. Although, ordinary labels are more informative than complementary labels and thus an assisted approach is needed to learn from. This paper uses chest X-ray because it is an image dataset that need time and cost for the labeling. This paper tries to classify normal, viral pneumonia and COVID-19 images. It uses linear/multilayer perceptron models for complementary label and use ResNet-50 model for ordinary label. It also checks the result with ordinary labels and compares its accuracy with complementary labels. With ordinary labels, it gets accuracy around 96,97%. After implementing it with complementary labels, it gets accuracy around 88%.