Software Defect Categorization based on Maintenance Effort and Change Impact using Multinomial Naïve Bayes Algorithm

Abstract

Cost-to-fix software defects increase exponentially as the software moves forward in its life cycle. The fixing of defects found during the maintenance phase is very costly. Software Defect Categorization (SDC) models categorize the software defects into different levels (high, moderate and low) based on various attributes. Such models can aid the software maintenance team to prioritize the defects and plan the resources effectively in fixing them. This study develops SDC models based on three software bug attributes-i) maintenance effort needed to fix a defect ii) change impact on the software for fixing a defect and iii) the combined approach of both- maintenance effort and change impact. In this study, we extracted the important features from the defect reports using text mining and the classification models are developed for each of the attributes using Multinomial Naïve Bayes (NBM) algorithm. The capability of the models is calculated using Area Under the ROC (Receiver Operating Characteristic) curve (AOC). The results of the study show that i) SDC models based on maintenance effort and change impact are capable to categorize the software defects ii) the performance of the SDC models based on the combined approach is better than SDC models based on maintenance effort and change impact.