IPSJ Digital Courier
Online ISSN : 1349-7456
ISSN-L : 1349-7456
Mining Quantitative Rules in a Software Project Data Set
Shuji MorisakiAkito MondenHaruaki TamadaTomoko MatsumuraKen-ichi Matsumoto
Author information
JOURNAL FREE ACCESS

2007 Volume 3 Pages 518-527

Details
Abstract

This paper proposes a method to mine rules from a software project data set that contains a number of quantitative attributes such as staff months and SLOC. The proposed method extends conventional association analysis methods to treat quantitative variables in two ways: (1) the distribution of a given quantitative variable is described in the consequent part of a rule by its mean value and standard deviation so that conditions producing the distinctive distributions can be discovered. To discover optimized conditions, (2) quantitative values appearing in the antecedent part of a rule are divided into contiguous fine-grained partitions in preprocessing, then rules are merged after mining so that adjacent partitions are combined. The paper also describes a case study using the proposed method on a software project data set collected by Nihon Unisys Ltd. In this case, the method mined rules that can be used for better planning and estimation of the integration and system testing phases, along with criteria or standards that help with planning of outsourcing resources.

Content from these authors
© 2007 by the Information Processing Society of Japan
Previous article Next article
feedback
Top