An assessment and comparison of common software cost estimation modeling techniques

摘要: This paper investigates two essential questions related to data-driven, software cost modeling: (1) What modeling techniques are likely yield more accurate results when using typical development data? and (2) the benefits drawbacks of organization-specific data as compared multi-organization databases? The former question is important in guiding analysts their choice right type technique, if at all possible. In order address this issue, we assess compare a selection common fulfilling number criteria large multi-organizational database business application domain. Namely, these are: ordinary least squares regression, stepwise ANOVA, CART, analogy. latter feasibility databases build models gained from local, company-specific collection modeling. As subset multi-company came one organization, were able investigate issue by comparing with based on data. Results show that performances considered not significantly different, exception analogy-based which appear be less accurate. Surprisingly, standard factors (e.g., COCOMO-like factors, Function Points), organization specific did better than generic, models.