Assessing Case Base Quality - UKCBR 2005
by Rahul Premraj, Martin Shepperd

Miltos Petridis (Ed.), 10th UK Workshop on Case-based Reasoning, December 2005.

Download as PDF file.

Abstract

In recent years, case-based prediction has become a widely advocated technique for software cost estimation. Typically, such approaches are in essence k nearest neighbour methods supported by a case base of a feature vector per software project. Given that software project data is relatively rare ┬ľ data sets may contain as few as 20 cases ┬ľ it is common to find a relatively undiscriminating approach to the projects contained within the case bases. We hypothesise that meta-data generated by monitoring case performance can contribute to identifying misleading cases and improve predictions. This paper reports results of a pilot study in which we enriched our case base with meta-data to record performance behaviour of individual data sets. An external fuzzy model was used to classify individual cases as fit or unfit for future use. Misleading cases i.e. with poor predictive ability were seeded into the data set to assess the potential of the approach. Our results show that the model successfully identified the seed cases and refrained from using them during future retrievals.

BibTeX Entry

@inproceedings{premraj-ukcbr-2005,
    title = "Assessing Case Base Quality",
    author = "Rahul Premraj and Martin Shepperd",
    year = "2005",
    month = dec,
    booktitle = "10th UK Workshop on Case-based Reasoning",
    editors = "Miltos Petridis",
    location = "Cambridge, UK",
}

Show all publications of the Software Engineering Chair.