Classifier Ensemble Method Based On Confusion Entropy And Its Applicatio

Classification is a common task in the fields of machine learning and data mining.The traditional approach of machine learning is to find a classifier that is closest to the actual classification function.Integrated learning is to fuse the results of multiple learners and make overall predictions,which can achieve more significant generalization performance than a single learner and effectively improve the classification performance.This paper first summarizes the research status of classifier ensemble methods at home and abroad,and introduces the basic principles of classifier ensemble methods based on confusion entropy and their advantages in practical applications.Confusion Entropy(CEN)is a new classifier performance measure that is more discriminative than accuracy(ACC)and relative classifier information(RCI).Therefore,a classifier integration method based on confusion entropy(or EABC,for short)is proposed.This method sorts the candidate classifiers in ascending order according to their CEN values,traverses and selects the classifiers that can reduce the CEN value in order,so that each selection is superior to the previous combination,in order to obtain better integrated classification results.Experimental results using seven candidate classifiers on five UCI datasets show that the proposed method improves the classification performance compared with two classic classifier integration methods,Ada Boost and XGBoost,and three literature methods,thereby verifying the feasibility of the integration method.Now we have entered the post epidemic era,but most of the detection of COVID-19 is still using RT-PCR detection method,which cannot be quickly detected.In order to improve the detection efficiency and adapt to the development of the times,this paper fuses the proposed integration model with the convolutional neural network(CNN)model,and designs an integration model for COVID-19 detection based on CT images.First,the patient CT image dataset was preprocessed,and then CNN was used to extract features from the preprocessed dataset.Subsequently,five candidate classifiers were used for integration classification experiments.The experimental results of the proposed method are compared with several literature methods in terms of performance indicators,and the accuracy rate has been improved by 2 to 6percentage points,especially the highest recall rate by 10 percentage points.It can be seen that the detection method proposed in this article is superior to existing models and achieves the expected results.