An Experimental Study on The Effect of Resampling Techniques in Multiclass Imbalanced Data in Learning Sector

Abstract

Data skewness is considered as one of the pain points in machine learning. This can be found in all sectors of real life. In the present era all the educational institutions are trying to improve the quality of their education by reducing the failures and improving the performance of students. Learning analytics sector is having a large amount of data which is skewed. The spotting of minority class is pivotal in skewed data. This gives rise to a big summon for machine learning algorithms as they take on the data as balanced distribution of classes. Research is going on this field and a lot of methods are available to deal with imbalanced data. Resampling the dataset during pre-processing stage is one of the methods. Oversampling and undersampling are the techniques available with resampling. This work concentrates on an investigational study on the effect of resampling techniques like Random undersampling(RUS), Random oversampling(ROS), Synthetic Minority Oversampling Technique(SMOTE) and ADAptiveSYNthetic sampling(ADASYN) on multiclass imbalanced data in learning sector.