Towards ensemble-based use case point prediction

Abstract

Early-stage software effort estimation (SEE) is crucial for successfully completing any software project since it helps in project bidding and efficient resource allocation. Most SEE models consider software size as a key metric for estimating effort. Consequently, software size becomes vital for early-stage SEE. Recently, use case points (UCP), derived from use case diagrams, gained popularity among the research community. The researchers used different classical and learning models for UCP prediction. Although learning models performed better than the classical models, it is difficult to conclude which learning model is superior. Ensembling is considered one probable solution when the individual models are not performing well. However, the ensemble models are not explored for UCP prediction till now. Motivated by this, the current work presents an ensemble-based framework for UCP prediction and investigates different ensemble models. We conducted an experimental analysis over two publicly available UCP estimation datasets by implementing different ensemble models. The results show that the ensemble models outperformed the base learners used in this work. Further, we compared the best performing ensemble learner with the existing UCP prediction models in the literature and found an improvement in UCP prediction performance.