FEPP: Advancing Software Risk Prediction in Requirements Engineering Through Innovative Rule Extraction and Multi-Class Integration

Abstract

The increasing complexity of software projects makes it difficult to predict risks in software requirements, which is a crucial and essential part of the Software Development Life Cycle (SDLC). The failure of a software project may occur from an inability to appropriately anticipate such risks. Because it is the first stage of any software project, risk prediction has a greater significance in software requirements. Thus, ForExPlusPlus (FEPP), a novel model for risk prediction in software requirements, is proposed in this work. Standard models such as K-nearest Neighbor (KNN), Naïve Bayes (NB), Logistic Model Tree (LMT), Random Forest (RF), and Support Vector Machine (SVM) are used to benchmark the suggested model. The dataset from the Zenodo repository is used to train these models, and standard assessment criteria are used to evaluate the results. The accuracy analysis of the models is assessed critically using the precision, F-measure (FM), and Mathew's correlation coefficient (MCC), as well as the error rate using the Kappa Statistic (KS) and Mean Absolute Error (MAE). The suggested FEPP performs better overall, with an accuracy of 96.84%, whereas KNN performs the worst, with an accuracy of 50.99%.