An empirical comparison of machine learning algorithms for classification of software requirements

Abstract

Intelligent software engineering has emerged in recent years to address some difficult problems in requirements engineering. Requirements are crucial for software development. Moreover, the classification of natural language user requirements into functional and non-functional requirements is a fundamental challenge as it defines the fulfillment criteria of the users' expected needs and wants. Therefore the research of this article aims to explore and compare random forest algorithm and gradient boosting algorithm to determine the accuracy of functional requirements and non-functional requirements in the process of requirements classification through the conduct of experiments. Random forest and gradient boosting are ensemble algorithms in machine learning that combines the decisions from several base models to improve the prediction performance. Experimental results show that the gradient boosting algorithm yields improved prediction performance when classifying non-functional requirements, in comparison to the random forest algorithm. However, the random forest algorithm is more accurate to classify functional requirements.