Research On Classification Method For Imbalanced Data And Its Application

Imbalance learning refers to the situation that the number of training samples in different categories varies greatly in the classification task.In the past decade,this field has attracted more and more attention from researchers.In the past five years,it has shown explosive growth.Different from the traditional classification tasks,on the one hand,the data imbalance will lead to the performance of the traditional classification algorithm greatly reduced,on the other hand,the unbalanced learning often has the problem of the unequal cost of misclassification,so the research on imbalance learning is of great significance.Previous studies have proposed a large number of imbalance learning algorithms,but there is little research on improving the classification accuracy of fewer classes from the perspective of post-processing prediction probability.This thesis analyzes the relevant background knowledge of imbalance learning and the causes of data unbalanced.Based on theory and experiments,it discusses the influencing factors of imbalance learning and explains the current solutions of imbalance learning.Based on the current research,aiming at the problem of binary classifications and multi-classification,the paper improves the problem of low precision and cost sensitivity of a few kinds of samples in imbalance learning.This work optimizes the classification of unbalanced data through a two-stage learning process.In the first stage,the prediction probability is obtained,and in the second stage,the optimization algorithm is used to optimize the prediction probability to ensure better classification results.Experiments show that this method can improve the accuracy of a small number of samples on 10 real data sets with binary classifications,and has good practical value.In multi-classification,through the verification of artificial data sets and real data sets,this method has achieved higher accuracy on a small number of samples,and this method has been successfully applied to the prediction of farm exit The results show that the prediction accuracy is greatly improved.