Enhancing Software Defect Prediction Through Root Cause Analysis: A Hybrid Approach Integrating Permutation Importance with XGBoost (PERMBoost)

Software Defect Prediction (SDP) plays an essential role in ensuring software quality and minimizing the costs associated with software failures. Conventional defect prediction approaches often focus on a singular means of defect assessment, neglecting the complex nature of the problem. This paper introduces a hybrid Perm-Boost model for SDP, using strengths of permutation importance and the XGBoost algorithm for root cause analysis (RCA). Evaluated on real-world datasets from NASA-MDP and AEEEM repositories, Perm-Boost outperformed traditional methods, achieving F-m scores of
$99.01,99.19$
, and 79.91 on CM1, PC2, and
$EQ$
datasets, respectively. It demonstrated high precision
$(98.00\%$
for
$\text{CM}1,98.18\%$
for PC 2) and recall
$(97.01\%$
for CM1,
$99.00\%$
for PC2), surpassing conventional approaches. Additionally, the incorporation of permutation importance facilitated the clarification of the interconnection between software defects and their root causes. This insight enables developers to prioritize tasks based on the severity of identified issues. This paper contributes a hybrid Perm-Boost model for SDP, an in-depth RCA utilizing influential features, a comprehensive case evaluation demonstrating superiority over alternative techniques, and practical guidance for implementing this approach in software projects.