A Class Balancing Methods Comparison in Software Requirement Classification Using a Support Vector Machine

Fachrul Pralienka Bani Muhamad, Esti Mulyani, Munengsih Bunga, Achmad Mushafa

2022

Abstract

Miss analysis of software requirements in the development stage whether functional or non-functional leads to a significant impact on the quality derivation and cost & time escalation. Especially in agile approaches, such as scrum, some non-functional requirements often go unnoticed, because of a high focus on business functionality that tends to be prioritized. Previous research has been carried out in classifying software requirements, especially non-functional requirements using the PROMISE dataset with the Bag of Words (BoW) feature extraction and the Support Vector Machine (SVM) algorithm. The results obtained from the combination of these methods provide a better accuracy value than the combination of feature extraction and other classification algorithms. However, the software requirement dataset tends to be imbalanced considering that there are several non-functional requirements types, so the data number of each class might differ. In another study, it was stated that the imbalance of datasets could give not optimal classification results, so it is necessary to balance the data. This research proposes class balancing on the dataset after the feature extraction is carried out. The output of the balanced class is used for the classification process. The PURE dataset is used in this research considering that the dataset is open to researchers. After experimenting with the combination of BoW feature extraction, as well as class balancing methods (i.e. SMOTE, Borderline SMOTE, and SVM SMOTE), and classified using the SVM algorithm, it was found that BoW with SVM SMOTE produces the best value average with an accuracy of 78.7%, precision of 80.2%, recall of 78.7%, and F1-Score of 78.9. It has higher results than software classification without a class balancing in enhancement average value accuracy of 0.03%, precision of 0.05%, recall of 0.03%, and F1-Score of 0.04%.

Paper Citation

in Harvard Style

Pralienka Bani Muhamad F., Mulyani E., Bunga M. and Mushafa A. (2022). A Class Balancing Methods Comparison in Software Requirement Classification Using a Support Vector Machine. In Proceedings of the 5th International Conference on Applied Science and Technology on Engineering Science - Volume 1: iCAST-ES; ISBN 978-989-758-619-4, SciTePress, pages 366-369. DOI: 10.5220/0011803100003575

in Bibtex Style

@conference{icast-es22,
author={Fachrul Pralienka Bani Muhamad and Esti Mulyani and Munengsih Bunga and Achmad Mushafa},
title={A Class Balancing Methods Comparison in Software Requirement Classification Using a Support Vector Machine},
booktitle={Proceedings of the 5th International Conference on Applied Science and Technology on Engineering Science - Volume 1: iCAST-ES},
year={2022},
pages={366-369},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011803100003575},
isbn={978-989-758-619-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 5th International Conference on Applied Science and Technology on Engineering Science - Volume 1: iCAST-ES
TI - A Class Balancing Methods Comparison in Software Requirement Classification Using a Support Vector Machine
SN - 978-989-758-619-4
AU - Pralienka Bani Muhamad F.
AU - Mulyani E.
AU - Bunga M.
AU - Mushafa A.
PY - 2022
SP - 366
EP - 369
DO - 10.5220/0011803100003575
PB - SciTePress