Classification of functional and nonfunctional requirements based on convolutional neural network with flower pollination optimizer

Accurate and efficient classification of software requirements is crucial for the success of system, as functional and nonfunctional requirements define the fundamental characteristics and constraints of a system. Documenting software requirements in natural language can lead to uncertainties such as ambiguity, inconsistency, or poor readability. Additionally, manual extraction of requirements can be tedious and prone to errors due to the need for precise interpretation, which increases the risk of miscommunication and mistakes during the development process. So, it is vital to use effective techniques like natural language processing to clearly understand the requirements. This paper proposes the use of deep learning models in conjunction with natural language processing, followed by flower pollination optimizer, to automate the classification task. The methodology leverages natural language processing to extract meaningful features used to train a convolutional neural network model. The convolution neural network model is enhanced using the flower pollination optimizer algorithm to ensure better convergence. The approach is implemented using an industry SmartNet dataset. To tackle the challenge of class imbalance, the synthetic minority oversampling technique is used. The approach is validated on both balanced and unbalanced datasets to demonstrate its effectiveness. Results show that the CNN-FPO framework performs exceptionally well, with accuracies ranging between 94.48% and 97.13% for balanced and 87.45% to 98% for unbalanced dataset.