Abstract Context In recent years, the multi-class classification of non-functional requirements has seen improvements through the use of Machine Learning algorithms. However, challenges such as data scarcity and class imbalance persist, particularly for languages other than English, such as Spanish. Objective This study aims to analyze the performance metrics of Machine Learning algorithms for classifying non-functional requirements translated into and originally written in Spanish. It evaluates the effectiveness of dataset balancing techniques and conducts cross-dataset validation to assess the generalizability of the models. Method A dataset balancing process was conducted using a combination of oversampling and undersampling techniques. Six algorithms were trained in two experiments using a hyperparameter tuning process, employing two different datasets: PROMISE_exp_translated and the newly PROMISE_exp_balanced . The best-performing models were further tested on unseen data to evaluate their generalizability. Results Logistic Regression and Naive Bayes demonstrated superior performance on the translated dataset, achieving f1-scores of 82% and 81%, respectively. Although overall performance decreased on the balanced dataset, specific underrepresented classes such as Portability and Fault Tolerance benefited from the balancing process. Conclusion Shallow Machine Learning algorithms are effective for classifying Spanish non-functional requirements, particularly when addressing data imbalance. The study highlights the importance of dataset balancing in improving classification performance for specific classes and provides insights into the challenges of generalizing models across datasets.

Background: Ensemble models play an important role in integrating multiple classifiers in a wide range of applications, such as medical diagnosis, sentiment analysis, and financial market trends. In Requirements Engineering (RE), automatic requirements classification can be improved by the utilization of these models. Aims: This paper analyses the performance metrics of voting and stacking ensemble models for requirements classification prediction. Moreover, a cross-dataset validation was performed for the meta-models generated using the stacking ensemble method. Methods: Some previously trained base models and two datasets of software requirements written in Spanish (translated PROMISE_exp and ReSpa dataset) were used to build the ensemble models. Results: The results indicate that the stacking model achieved a weighted F1-score of 0.828 using Support Vector Machine (SVM) and Multi-layer Perceptron (MLP) for translated PROMISE_exp dataset. For the ReSpa dataset, the stacking model achieved a weighted F1-score of 0.890 using Logistic Regression (LR). Conclusion: This study confirms a slight improvement in the performance of binary requirements classification using stacking ensemble methods over voting and most individual base models. Moreover, combining all models outperforms combinations that include only Shallow ML or DL models.

Machine Learning (ML) algorithms have become a powerful instrument in software requirements classification. Nevertheless, most of the research focusing on requirements is in English, with less attention to other languages. Given a lack of datasets in Spanish, we created a new dataset from a collection of requirements from final degree projects from the University of A Coruña. In this paper, we investigate which combinations of text vectorization techniques with ML algorithms perform best for requirements classification in a Spanish dataset. We found that SVM with TF-IDF gives the highest f1-score (0.95 and 0.79 for functional and non-functional classification).

Traditional methods for requirements prioritization (RP) are currently limited by scalability and lack of automation issues. In recent years, there has been an exponential growth in the use of Artificial Intelligence (AI) techniques in different areas of software engineering (e.g., requirements analysis, testing, maintenance). In particular, I have found thirteen RP methods applying AI techniques such as machine learning, or genetic algorithms. 38% of these approaches seek to improve the scalability problem, whereas only 15% of them aim to improve the automation aspect along the RP process. Moreover, all these studies have carried out their evaluations with a number of requirements no greater than 100.In order to address the issues of scalability and lack of automation in RP, the present research project aims to propose a semi-automatic multiple-criteria prioritization method for functional and non-functional requirements of software projects developed within the Software Product-Lines paradigm. The proposed RP method will be based on the combination of Natural Language Processing techniques and Machine Learning algorithms, and for its validation, empirical studies will be carried out with real web-based geographic information systems (GIS). This paper describes the problem and technical challenges to be addressed, the related works, as well as the main contributions of the proposed solution.