AODE for Source Code Metrics for Improved Software Maintainability

TL;DR: A novel classification method--Aggregating One-Dependence Estimators (AODE) is proposed to support and enhance the understanding of software metrics and their relationship to software quality and a Symmetrical Uncertainty based feature selection method is presented.

Abstract: Software metrics are collected at various phases of the whole software development process, in order to assist in monitoring and controlling the software quality. However, software quality control is complicated, because of the complex relationship between these metrics and the attributes of a software development process. To solve this problem, many excellent techniques have been introduced into software maintainability domain. In this paper, we propose a novel classification method--Aggregating One-Dependence Estimators (AODE) to support and enhance our understanding of software metrics and their relationship to software quality. Experiments show that performance of AODE is much better than eight traditional classification methods and it is a promising method for software quality prediction. Furthermore, we present a Symmetrical Uncertainty (SU) based feature selection method to reduce source code metrics taking part in classification, make these classifiers more efficient and keep their performances not undermined meanwhile. Our empirical study shows the promising capability of SU for selecting relevant metrics and preserving original performances of the classifiers.