Simplifying Software Metric Models via Hierarchical LASSO with Incomplete Data Samples

Software metric models can be used in predicting the interested target software metric(s) for future software project based on certain related metric(s). However, during the construction of such a model, incomplete data often appear in data sample gained from analogous past projects. In addition, whether a particular continuous predictor metric or a particular category for a certain categorical predictor metric should be included in the model must be determined in practice. To solve these problems, this paper introduces a methodology integrating the k-nearest neighbors (k-NN) multiple imputation method, kernel smoothing, Monte Carlo simulation, and a latest variable selection method. Thus, a more flexible model is constructed. A case study is given to illustrate the proposed procedures.

I. INTRODUCTION

Software metrics are measurements or indications of some properties of a software system, its specifications, and/or the software project that developed it. Typical examples are the program size of a system (e.g., in number of lines of code), the number of defects in a system found during initial live-run, the project team size, the project work effort (e.g., in number of person-hours), etc. [1].