Uma metodologia baseada em modelos estatísticos e redes neurais para a estimativa de esforço de desenvolvimento de software / A methodology based on statistics models and neural networks to the software project devolopment effort estimation

AUTOR(ES)
DATA DE PUBLICAÇÃO

2007

RESUMO

Software effort estimation is an important part of software development work and provides essential input to project feasibility analyses, bidding, budgeting and planning. The consequences of inaccurate estimates can be severe. Optimistic estimates may cause significant losses while the pessimistic estimates may lead to loss of exiting and future contracts. Unfortunately, it is common for software development projects to overrun their effort estimates, typically because the estimates are too optimistic. This thesis presents a methodology based on statistical and neural networks methods to provide more accurate effort estimates in a simpler way. The goal of this research is to contribute to reduce estimation error in software development projects by better understanding the different software effort estimation models and techniques that include: artificial neural networks, case-based reasoning techniques, regression-based models, and techniques for integrating analysis of residuals, analysis of variance and regression-based models. Several case studies have been conducted. The results show all the proposed models lead to realistic estimations, however, neural networks based models emerge as a very easy tool for local models calibration processes due to its simpler implementation. The case studies show all the models are sensitive to the available data, thus requiring recalibration processes every since new project data area gathered to the database.

ASSUNTO(S)

estimativa de esforço de software analysis of variance computação aplicada redes neurais artificais artificial neural networks regression analysis análise de variância software effort estimation análise de resíduos analysis of residuals análise de regressão

Documentos Relacionados