Qorry Meidianingsih

Program Studi Pendidikan Matematika, Universitas Negeri Jakarta, Jakarta Timur

Devi Eka Wardani Meganingtyas

Program Studi Matematika, Universitas Negeri Jakarta, Jakarta Timur

Sudarwanto

Program Studi Matematika, Universitas Negeri Jakarta, Jakarta Timur

Dian Handayani

Program Studi Statistika, Universitas Negeri Jakarta, Jakarta Timur

Pinta Deniyanti Sampoerno

Program Studi Pendidikan Matematika, Universitas Negeri Jakarta, Jakarta Timur

Makmuri

Program Studi Pendidikan Matematika, Universitas Negeri Jakarta, Jakarta Timur

DOI: https://doi.org/10.19184/mims.v26i1.60019

ABSTRACT 

The study aims to compare the performance of the Support Vector Machine (SVM) classification method under several approaches for handling imbalanced class data. The methods considered in this study include Underbagging, SMOTEBagging, Safe-Level SMOTEBagging, and K-Means clustering-based undersampling. Eight datasets obtained from the UCI Machine Learning Repository, exhibiting varying degrees of class imbalance, were employed in the experiments. Overall, the accuracy values achieved by the four imbalance-handling approaches are generally satisfactory. However, SMOTEBagging and Safe-Level SMOTEBagging consistently yield the highest accuracy performance, whereas the K-Means-based approach produces the lowest average accuracy. In terms of sensitivity, all approaches demonstrate comparable performance. A similar pattern is observed for specificity, which aligns with the findings obtained from the accuracy metric. Considering the evaluation based on accuracy, sensitivity, and specificity, all four approaches can be regarded as effective in handling class imbalance. Nevertheless, when assessing the consistency across these three performance measures, ensemble learning-based methods, particularly Safe-Level SMOTE Bagging,exhibit superior and more stable performance compared to the other approaches.

Keywords: class imbalance, undersampling, ensemble method, clustering, support vector machine.

MSC2020: 62-08