Investigating effort prediction of web-based applications using CBR on the ISBSG dataset

As web-based applications become more popular and more sophisticated, so does the requirement for early accurate estimates of the effort required to build such systems. Case-based reasoning (CBR) has been shown to be a reasonably effective estimation strategy, although it has not been widely explored in the context of web applications. This paper reports on a study carried out on a subset of the ISBSG dataset to examine the optimal number of analogies that should be used in making a prediction. The results show that it is not possible to select such a value with confidence, and that, in common with other findings in different domains, the effectiveness of CBR is hampered by other factors including the characteristics of the underlying dataset (such as the spread of data and presence of outliers) and the calculation employed to evaluate the distance function (in particular, the treatment of numeric and categorical data).