Automated Classification of User Requirements written in Arabic using machine learning algorithms

Authors

Ahmad Althunibat, Faculty of Sciences and Information Technology, Al-Zaytoonah University of Jordan, Amman, JordanFollow
Reem Amro, Faculty of Sciences and Information Technology, Al-Zaytoonah University of Jordan, Amman, JordanFollow
Belal Hawashin, Faculty of Sciences and Information Technology, Al-Zaytoonah University of Jordan, Amman, JordanFollow
Hend AlNuhait, Faculty of Computer Studies, Arab Open University, Riyadh, Saudi ArabiaFollow
Sally Almanasra, Faculty of Computer Studies, Arab Open University, Riyadh, Saudi ArabiaFollow
Haneen A. Al-Khawaja, Faculty of Business, Amman Arab University, Amman, Jordan\\Swiss FinTech Innovation Lab, University of Zurich, SwitzerlandFollow

Author Country (or Countries)

Switzerland

Abstract

Requirement engineering is a critical step in software engineering, influencing software development outcomes. The manual classification of software requirements into Functional Requirements and Non-Functional Requirements is a laborious and costly process with varying accuracy. Errors in classification can lead to misunderstandings and incomplete products. Arabic requirements pose additional challenges due to their structural and semantic characteristics, contributing to inherent ambiguity. In addition to the lack of Arabic Studies as well as the public datasets for requirements written in Arabic.This study proposes combining machine learning and deep learning algorithms, including K-nearest neighbors (KNN), Support Vector Machines (SVMs), Random Forest, XGBoost, and the Arabert model, with optimization techniques to streamline the categorization of Arabic requirements. Optimal configurations for several classifiers are identified and examined by stemming techniques. In addition, an Arabic data set for requirements was collected. The results demonstrate the effectiveness of the proposed approach, enhancing productivity and mitigating risks. The SVM classifier achieves an F1-Score of 0.93, while combining it with ISRIStemmer improves the score to 0.95. The Arabert model achieves the highest F1-Score of 0.97, highlighting its performance in classifying Arabic requirements accurately.

Digital Object Identifier (DOI)

https://dx.doi.org/10.18576/amis/170620

Recommended Citation

Althunibat, Ahmad; Amro, Reem; Hawashin, Belal; AlNuhait, Hend; Almanasra, Sally; and A. Al-Khawaja, Haneen (2023) "Automated Classification of User Requirements written in Arabic using machine learning algorithms," Applied Mathematics & Information Sciences: Vol. 17: Iss. 6, Article 26.
DOI: https://dx.doi.org/10.18576/amis/170620
Available at: https://digitalcommons.aaru.edu.jo/amis/vol17/iss6/26