Improving Prediction Models of Student Business Career Using Sampling Techniques for Learning in Multi-class Imbalance Data Set

Abstract

The purposes of the research  are  1)  to study the sampling techniques for learning in multiclass imbalanced datasets   and    2)  to compare the efficiency of the sampling techniques for learning in multiclass imbalanced datasets.  The WEKA program is used to adjust the balance by the techniques of Over-sampling, Under-sampling and SMOTE. The prediction models are assessed  by the techniques of Decision Tree and Random Forest to compare the highest forecasting result.

The results of the research indicate that the datasets  balanced by Over-sampling and compared by Decision Tree and Random Forest  techniques show that Random Forest is the  most efficient with accuracy 67.17 %, precision 0.66 %, recall 0.67 %  and  F-measure 0.66 %.