A Classification Method For Imbalanced Data Based On SVM

Due to imbalanced data in the real case is becoming more common,therefore be-come a hot issue in the field of data mining.However our existing traditional classifica-tion algorithms' classification performance is not ideal when doing with the imbalanced data classification problem,the main defect is always misjudging minority class samples to majority class.But in most practical problems,Minority class samples is our really concern.So,the study of imbalanced data classification problem,especially improving the ability of the recognition of the minority class samples,not only has theoretical significance,but more practical significance.To solve the problem of imbalanced data classification,there are two main research directions:the data level and algorithmic level.This article obtains from the data level with the new algorithm NSSMOTE,reshaping the dataset through synthetic more minority class samples.This method is based on the SVM classification results,for the first time,according to the SVM separation hyperplane,we can get the resulting distance of each sample to the separating hyperplane,then according to the distance to get each sample different weights;Refer to SMOTE of the thought of stochastic linear interpolation,pick K minority class samples randomly,work out their weighted sum to get the new synthetic minority sample.Experiments show that the data after processed by NSSMOTE algorithm is better compared to the original data,SMOTE,ROS algorithm,BSMOTE algorithm,and ADASYN algorithm classification performance as for the indicators of F1 and G-mean.