Non-functional Requirements Classification using Artificial Neural Networks

Requirements classification is a task commonly made by the human. This fact makes the process error-prone and expensive in a matter of time and effort. This study aims to classify non-functional requirements using a Shallow Artificial Neural Network to support the requirements classification while analyzing its architectural features. We used an existing preprocessed dataset to compare the results. The non-functional requirements classified were: availability, failure tolerance, maintainability, performance, scalability, security, and usability. A set of experiments were performed to adjust and analyze the features of the Artificial Neural Network: number of neurons, activation functions, additional hidden layer, and learning rate. The obtained neural network, which resulted in a simpler architecture, outperforms the results reported in the literature for that particular dataset.