Performance Comparison of various T-Norms with Choquet integral in Fuzzy Logic using Covid-19 Chest X-Ray Data

Abstract

Covid-19 declared by WHO as a global pandemic leading to millions of deaths. In this paper we are integrating various t-norms in fuzzy logic and want to see which t-norm will perform the best. Here the ensemble model is used to distinguish chest X ray images as Covid infected, Pneumonia infected and Normal patients. Transfer learning technique is used to train four very powerful CNN’s namely Vgg16, Restnet50, InceptionnetV3 and Densenet121.These pretrained CNN models are feature extracted and then used as classifiers for the chest X- ray images. After that the prediction results of the individual models are aggregated using Choquet integral with various t-norms based fuzzy measure like Lukaseiweiz , Hemaecher and Nilpotent and the final labels are predicted which are more strong than the prediction of individual models. In order to evaluate the proposed model chest x ray images from public repositories like IEEE and Kaggle are used. The final prediction results are better than the individual results of the base CNN models.