Classification and Prioritisation of Software Requirements using Machine Learning – A Systematic Review

Requirement Engineering (RE) plays an integral role throughout the process of software development. Requirement identification and prioritisation are the foremost phases of the RE process. Latest RE research work uses Machine Learning (ML) algorithms to tackle RE problems such as identifying requirements and assigning priorities to requirements, which have given better results than that of traditional natural language processing methods. An adequate understanding of these ML methods, however, is still lacking. The aim of this study is to understand which of the ML algorithms is likely to classify and prioritise the requirements efficiently and how they can be evaluated. It is observed that the current approaches are having constraints of scalability and complexity. Different methods used for the text preprocessing of requirements from SRS and user reviews are also proposed. 6 different ML algorithms and 6 different prioritisation algorithms, which are most common methods, are found. The most popular performance parameters used are accuracy, precision and recall. The limitations of these ML approaches are irrespective of dependency of requirements, priorities are assigned to requirements, the results with respect to scalability and speed is inferior.