FEPP: MULTI-CLASS SOFTWARE RISK PREDICTION IN REQUIREMENTS ENGINEERING VIA RULE EXTRACTION

Abstract

Requirements Engineering (RE) is an urgent phase in software engineering where vague, incomplete, and inconsistent requirements result in many project risks such as delay, cost overruns, and outright failure. Most of the present risk prediction models adhere to binary classification and thus are unable to address the complex and diverse reality of risk scenarios. This paper presents Feature-Enriched Prediction Paradigm (FEPP)—a new framework that combines rule extraction techniques with multi-class classification to predict risks during the RE phase more effectively. FEPP uses Association Rule Mining (ARM) and Fuzzy Logic mechanisms for dynamic rule extraction from historical project log data to identify and predict future risk patterns. To increase the prediction accuracy, FEPP combines various classifiers including Random-Forest, XGBoost, and Gradient-Boosting through the Voting Classifier Mechanism (VCM). Experimental results on benchmark RE datasets show that FEPP achieves 12–15% more accurate prediction over traditional models, facilitating early risk mitigation and improved project outcomes.