Working Paper Details

Cluster Specific Software Effort Estimation for Multi-Company Datasets
Theodoros Evgeniou and Kishore Sengupta


Predicting accurately the required effort for software projects is critical in software engineering. To this end predictive models estimated from data describing past projects can be used. When a company does not have enough data to develop its own company specific models, it can use generic models estimated from multi-company datasets. However, experiments show that such generic multi-company models have high predictive errors and perform worse than company specific models developed in the case a company has enough data. Instead of either company specific or generic multi-company models, in this paper the authors develop cluster specific models from a multi-company dataset following a novel methodology using clustering and classification techniques. They compare the new models with generic multi-company as well as company specific ones. They show that they are better than the first, and, although the particular models they develop are worse than the second, they show evidence that it is possible to develop cluster specific models that are also better than the company specific ones. They also present a novel methodology, based on classification tools such as decision trees, to finding what are the important factors influencing productivity for a particular software project. They test their methods using the widely used European Space Agency database of software projects.

- back -
Share this page
Quick Links

INSEAD Case Distributors

INSEAD Knowledge

INSEAD Libraries



Executive MBA

Executive Education