A Framework to Automate Requirements Specification Task

Abstract

Requirement identification and prioritisation are two principal activities of the requirement engineering process in the Software Development Life cycle. There are several approaches to prioritization of requirements identified by the stakeholders. However, there is a need for a deeper understanding of the optimal approach. Much study has been done and machine learning has proven to help automate requirement engineering tasks. A framework that identifies the types of requirements and assigns the priority to requirements does not exist. This study examines the behaviour of the different machine learning algorithms used for software requirements identification and prioritisation. Due to variations in research methodologies and datasets, the results of various studies are inherently contradictory. A framework that identifies the types of requirements and assigns the priority to requirements does not exist. This paper further discusses a framework for text preparation of requirements stated in natural language, type identification and requirements prioritisation has been proposed and implemented. After analysing the ML algorithms that are now in use, it can be concluded that it is necessary to take into account the various types of requirements when dealing with the identification and classification of requirements. A Multiple Correlation Coefficient-based Decision Tree (MCC-based DT) algorithm considers multiple features to map to a requirement and hence overcomes the limitations of the existing machine learning algorithms. The results demonstrated that the MCC-based DT algorithm has enhanced type identification performance compared to existing ML methods. The MCC-based DT algorithm is 4.42% more accurate than the Decision Tree algorithm. This study also tries to determine an optimisation algorithm that is likely to prioritise software requirements and further evaluate the performance. The sparse matrix produced for the text dataset indicates that Adam optimisation method must be modified to assign the requirement a more precise priority. To address the limitations of the Adam Algorithm, the Automated Requirement Prioritisation Technique, an innovative algorithm, is implemented in this work. Testing the ARPT on 43 projects reveals that the mean squared error is reduced to 1.34 and the error cost is reduced to 0.0001. The results indicate an 84% improvement in the prioritisation of requirements compared to the Adam algorithm.