NLP Based Framework for Selecting SDLC Models in Software Development Process

Abstract

Requirement engineering plays a significant part in SDLC requirements. The classification of needs is one of the most important aspects of the requirement engineering process. Requirements classification can be done manually, but it takes a lot of time, effort, money, as well as varying accuracy. Within software engineering, proper requirement classification became a necessary task, as many earlier studies have recommended automating the classification process, but the classification process automation was insufficient. We present a strategy to automatically categorize software requirements NLP that can support the SDLC and its selection. The PROMISE exp dataset, which includes labeled requirements, was used in this investigation. All software documents in the database were altered using a series of procedures. For classification, Word2Vec, Fast Text and BERT used the several different Machine learning algorithms. The PROMISE exp data, information about the processes required to re-perform the classification, and the Measurement were all used in this study. The classification of requirements using SVM and KNN algorithms can serve as a way and resource for another investigation, according to Bag of Words. This research compares text feature extraction techniques and NLP techniques in the context of requirements classification. The work is based on two major questions: which NLP approach works best for classification of Software Requirements into Functional and Non-Functional Requirements (NF), and the sub-categories of NFR's, and to determine which Machine Learning Algorithm gives the best results. That can in turn help us to predict the appropriate SDLC model. The PROMISE exp dataset, a freshly created dataset that expands the already known PROMISE repository, a repository that contains labeled software needs, was utilized to conduct the research. Non-functional and functional requirements for various types of software products are included in this dataset. All of the documents in the database were normalized, and Word2Vec, Fast-text and BERT were utilized as feature extraction and feature selection algorithms. Support Vector Machine (SVM), Decision Tree, Random Forest, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN) were the categorization techniques employed. The data used in the experiment, the intricacies of the processes used to reproduce the classification, and the comparison of word embedding techniques for this repository, which have not been covered by other research, are all novel aspects of our work. This study is based on automated classification of software requirements into functional and sub classes of non-functional requirements, it would act as a resource for the software industry and will aid other researchers in their understanding of the process of requirement classification and predict the best possible SDLC.