Enhancing Software Requirement Classification via Dataset Fusion and Machine Learning

Software engineering involves numerous steps; a successful software product follows these guidelines to the core. One such step is gathering requirements for a software product. This step is quite expensive in terms of time and money; a potential solution is to automate the requirement collection process. Automating the process of gathering software requirements requires separating requirements into types. An approach to predict the type of requirement is using text classification and machine learning; however, the problem with this approach is that it requires a large amount of data, which is not available for this use case. In this study, we perform dataset fusion to create a large dataset. We applied vertical fusion, which increased the number of instances in the dataset. Once a fusion-based dataset is created, machine learning algorithms are applied, and based on empirical results, the performance of the machine learning model after fusion drastically improved to 87.78% f1score with support vector machine (SVM). This improvement shows the efficacy of data fusion in improving the performance of a text classifier and demonstrates that it can overcome the limitations of small datasets by combining data from diverse sources. Our study demonstrated the robustness of our approach in software requirement classification by surpassing the highest recall scores from the previous four years, achieving 94.20% with fusion-based SVC and outperforming previous models even in non-fusion settings.