Software Requirements Classification using Machine Learning algorithm’s
Gaith Y Quba, Hadeel Al Qaisi, Ahmad Althunibat, Shadi AlZu’bi
Abstract
The world is growing and developing rapidly, and the demand for software has been increasing speedily, any software has many steps for building a program and all the steps are important for software requirements. Requirements classification can be applied manually, which requires great effort, time, cost and the accuracy may vary. Therefore, many previous researchv has been proposed to automate the classification process, but the automation process of the classification was not sufficient. In this study, we will propose a technique to automatically classify software requirements using machine learning to represent text data from software requirements specification and classify requirement to group Functional Requirement and Non-Functional Requirement. The experimented dataset in this study was the PROMISE_exp, which includes labeled requirements. All the documents of software from the database were changed (cleaned) with a set of steps (normalization, extractions, selection any techniques that will be used. The BoW used SVM algorithm or KNN algorithm for classification. This study used data from the PROMISE_exp to do the work, the information of the steps used to re-performed the classification, and the Measurement BoW, when using SVM and KNN algorithms the classification of requirements making can serve as a way and resources for another study. It can be seen that the use of BoW with SVM is better than use KNN algorithms with an average F-measure of all cases of 0.74. In future work we intend to improve to technique with make merge and change some algorithms as Logiest Regression to improve the Accuracy ( Precision) of our model.