Increasing the accuracy and reliability of analogy-based cost estimation with extensive project feature dimension weighting

Accurate and reliable software cost estimation is a vital task in software project portfolio decisions like resource scheduling or bidding. A prominent and transparent method of supporting estimators is analogy-based cost estimation, which is based on finding similar projects in historical portfolio data. However, the various project feature dimensions used to determine project analogy represent project aspects differing widely in their relevance; they are known to have varying impact on the analogies - and in turn on the overall estimation accuracy and reliability - , which is not addressed by traditional approaches. This paper (a) proposes an improved analogy-based approach based on extensive dimension weighting, and (ii) empirically evaluates the accuracy and reliability improvements in the context of five real-world portfolio data sets. Main results are accuracy and reliability improvements for all analyzed portfolios and quality measures. Furthermore, the approach indicates a quality barrier for analogy-based estimation approaches using the same basic assumptions and quality measures.