Software Effort Estimation: a Review of Taxonomies for use to-wards a Probabilistic Evidence Synthesis
This paper presents a systematic literature review of six Software Effort Estimation taxonomies, revealing significant structural fragmentation, a lack of coverage for emerging application domains like mobile and AI, and a general absence of external validation, thereby highlighting the urgent need for a unified, industry-validated taxonomy.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Every software project, from a simple mobile app to a complex banking system, begins with a question that is deceptively simple: how much work will this take? In the world of software engineering, this is known as effort estimation. It is the process of predicting the time and resources required to build a piece of software before a single line of code is written. This prediction is vital; it determines the budget, the schedule, and whether a project can be staffed with the right people. However, predicting the future of software is notoriously difficult because the factors that influence how long a project takes are vast, varied, and often uncertain. A project might run longer because the team is inexperienced, because the technology is new, or because the requirements keep changing. To make sense of this chaos, researchers have spent decades trying to organize these factors into structured lists, or taxonomies, which act as maps to help experts navigate the complex terrain of software development.
Despite the clear need for such maps, a new review of the field reveals that the existing maps are fragmented and incomplete. Emilia Mendes and Marc Solé Estragués, researchers at Aarhus University in Denmark, set out to examine every known classification system for software effort estimation. Their goal was to see if these systems could be combined to create a unified tool for the future, or if they were too different to be reconciled. They conducted a systematic review, a rigorous method of searching through scientific literature to find and analyze every relevant study that proposed a way to categorize the factors affecting software work. After sifting through thousands of articles from major academic databases, they found only six studies that actually proposed a taxonomy for this specific purpose. This small number itself was a significant finding, suggesting that while the problem is well-known, the effort to solve it has been sporadic and scattered over the last two decades.
The researchers then examined these six taxonomies in detail, looking at how they were built and what they contained. They found that the maps were drawn in very different styles. Some were broad, covering everything from the size of the software to the experience of the team and the specific techniques used to make the estimate. Others were narrow, focusing on just one aspect, such as only the size of the code or only the methods used to predict the work. The structures varied wildly, with some lists containing just a handful of categories and others sprawling into hundreds of specific details. When the researchers compared the lists side by side, looking for shared terms and concepts, they discovered a startling lack of agreement. Out of hundreds of specific factors listed across all six taxonomies, only a tiny fraction were described in the same way. Most of the terms were unique to a single study, meaning that two experts using different maps would be talking past each other, using different words for the same idea or different ideas for the same word.
The review also highlighted a significant gap in coverage. The existing maps focused heavily on traditional types of software, such as business applications and embedded systems found in devices. However, they largely ignored the rapidly growing fields of artificial intelligence software and modern mobile applications. Furthermore, the researchers found that most of these taxonomies had never been tested in the real world. They were proposed as ideas or theoretical frameworks, but there was little evidence that they actually helped practitioners in industry make better predictions. Only one of the six taxonomies had been applied in a real-world setting to see if it worked. The others remained unvalidated, leaving a question mark over their practical usefulness.
The authors argue that this fragmentation is a barrier to progress. When researchers and practitioners cannot agree on a common language, it becomes difficult to share knowledge, compare results, or build upon previous discoveries. The review suggests that the field needs to move beyond these isolated lists. The researchers propose that the next step is to either extend the existing maps to cover the missing areas of software development or to build a single, unified taxonomy that covers all types of software. Such a unified system would need to be tested in industry to ensure it is actually helpful. Until then, the field remains a collection of disconnected islands, where the potential for a clear, shared understanding of software effort is still waiting to be realized.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.