Automated identification of business rules in requirements documents

Business Rule identification is an important task of Requirements Engineering process. However, the task is challenging as business rules are often not explicitly stated in the requirements documents. In case business rules are explicit, they may not be atomic in nature or, may be vague. In this paper, we present an approach for identifying business rules in the available requirements documentation. We first identify various business rules categories and, then examine requirements documentation (including requirements specifications, domain knowledge documents, change request, request for proposal) for the presence of these rules. Our study aims at finding how effectively business rules can be identified and classified into one of the categories of business rules using machine learning algorithms. We report on the results of the experiments performed. Our observations indicate that in terms of overall result, support vector machine algorithm performed better than other classifiers. Random Forest algorithm had a higher precision than support vector machine algorithm but relatively low recall. Naïve Bayes algorithm had a higher recall than support vector machine. We also report on evaluation study of our requirements corpus using stop-words and stemming the requirements statements.