Classification Based Software Defect Prediction Model for Finance Software System - An Industry Study

Abstract

Automated software defect prediction is an important and fundamental activity in the domain of software development. Successful software defect prediction can save testing effort thus reduce the time and cost for software development. However, software systems for finance company are inherently large and complex with numerous interfaces with other systems. Thus, identifying and selecting a good model and a set of features is important but challenging problem. In our paper, we first define the problem we want to solve. Then we propose a prediction model based on binary classification and a set of novel features, which is more specific for finance software systems. We collected 15 months real production data and labelled it as our dataset. The experiment shows our model and features can give a better prediction accuracy for finance systems. In addition, we demonstrate how our prediction model helps improve our production quality further. Unlike other research papers, our proposal focuses to solve problem in real finance industry.