Locality preserving projection on source code metrics for improved software maintainability

Abstract

Software project managers commonly use various metrics to assist in the design, maintaining and implementation of large software systems. The ability to predict the quality of a software object can be viewed as a classification problem, where software metrics are the features and expert quality rankings the class labels. In this paper we propose a Gaussian Mixture Model (GMM) based method for software quality classification and use Locality Preserving Projection (LPP) to improve the classification performance. GMM is a generative model which defines the overall data set as a combination of several different Gaussian distributions. LPP is a dimensionality deduction algorithm which can preserve the distance between samples while projecting data to lower dimension. Empirical results on benchmark dataset show that the two methods are effective. © Springer-Verlag Berlin Heidelberg 2006.