Correlation of Traditional Technique and ML-Based Technique for Efficient Effort Estimation: In Agile Frameworks

Successful planning and monitoring of a project can be achieved from an accurate effort estimation during the initial phase of every sprint in Agile (Scrum). Project complexity, project cost, and resources are a few of the unknowns that a team tries to estimate in project planning sessions for efficient and quick completion of a product in a competitive environment. To estimate effort various traditional approaches are defined, but the accuracy of those estimates is still one of the major challenges the development team faces. According to the studies, the traditional approach is highly based on human-based guesstimate methods, while supervised Machine learning techniques such as Multilayer Perceptron (MLP) and Gradient Boosting (GBM) generated high accuracy with extremely low error rates for effort estimation. Various reviews in this matter are focused on either the Traditional approach or the Machine Learning (ML) approach but a concise comparison of accurate effort estimation is still missing in any recent literature. This paper focuses on the review of several studies and comes up with a comparison between traditional and ML approaches for accurate effort estimation. This paper also mentions several factors that could lower the accuracy for estimating effort even after applying ML such as limited dataset availability, use of irrelevant cost drivers in the dataset, lack of clarity in cost drivers, standardization anomaly for the use of evaluation metric, and ML technique. Finally, this paper is an ELR (Evidence-based Literature Review) via. which 3 research problems were addressed, offering valuable insights for future researchers.