Introduction
This is a project based on chest xray classifcation into three cases Covid-19, Pneumonia and Normal cases. For this purpose I have used three classifier models for this which are trained from scratch.
First Classifier
The first classifier is built using multiple convolutional layers which takes in a 3-channel RGB image to give the predictions.
Second Classifier
The second classifier is built using 4 convolutional layers.
Third Classifier
The third classifier only considers the red channel of the input image to give the prediction scores.
Sugeno Integral based Ensemble
In the case of a fuzzy inference system, output values are calculated based on some fuzzy rules andinput variables. These rules can be interpreted into two ways which are known as conjunctive rules anddisjunctive rules. Weighted maximum is a way of interpreting fuzzy inference rules in the conjunctivemanner, and weighted minimum is one way of interpreting fuzzy inference rules in a disjunctive manner.These two interpretations of fuzzy inference rules are unified by the Sugeno integral method usingweighted maximum and weighted minimum. Here, one important aspect of the weighted maximumand the weighted minimum used by Sugeno integral is that the weighting vector does not belong toany probability distribution, which means that the sum of the weights does not equal to 1. Hence, theaggregation operator that uses these weights does not satisfy the unanimity. Also, in the case of weightedmaximum and the weighted minimum, the values that need to be aggregated are independent of eachother which is not the case in the Sugeno integral method. So it can be said that the Sugeno integralcombines the conclusions from the different fuzzy inference rules, where the rules are not independent.As the weights are not probabilistic and need to be determined experimentally, so in order to apply theSugeno integral method to combine the predictions from three base learners, we have used optimization algorithms to get optimal fuzzy measures.
Choquet Integral based Ensemble
Aggregation refers to the process of collecting performance scores of each classifier into a single global score. The function that is used to combine the scores into a single global score is known as aggregation operator.Normally as the aggregation operator, we have weighted average or quasi arithmetic means butin our case it is the Choquet fuzzy integral operator. The Choquet fuzzy integral method has been used previously in many pattern recognition problems. The advantage of using it is thatit harnesses the degree of uncertainty that is present in the decision scores that we get as additionalinformation during the fusion of classifiers which is absent in normal ensembling methods. So the endresult is the generalization of aggregation operators on a set of confidence scores which are known asfuzzy measures. These fuzzy measures are determined using the optimization algorithms.
Research Publication
More better understanding of the ensemble methods used here please refer to data
Dataset Licence
The models have been trained using the dataset given in data that was released under a CC BY 4.0 licence.