An experimental analysis of traditional machine learning techniques for classification of non-functional requirements

Requirement analysis is critical step before the starting of software development. There are two kinds of software requirement in software development: (1) functional requirement and (2) non-functional requirement. Visibility of Functional Requirement (FR) is very much clear in software development, but Non-Functional Requirement (NFR) is hidden in nature so much less research has been done in the area of NFR. Even though it is hidden requirement it still plays a very important role in software development because it specifies quality and constraints of the system. To automate the process of requirement classification in software development, Machine Learning (ML) techniques are used. This paper presents the exhaustive experimental analysis of traditional ML techniques used for classification of NFR. In this experimental analysis Multinomial Naive Bayes (MNB), Logistic Regression (LR), Support Vector Machine with Stochastic Gradient Descent (SVM-SGD) and K-Nearest Neighbours (KNN) algorithms are analysed in terms of accuracy, recall, precision, and F1-Score to classify the NFR. As the data set to perform this experimental analysis, the PROMISE repository is used. In this work results show that SVM-SGD outperforms all the ML techniques by giving the F1-Score 00.92, Recall 00.92, Precision 00.93 and Accuracy 00.92.