Implementation of Chernobyl disaster optimizer based feature selection approach to predict software defects.

Area of Science:

• Software Engineering
• Machine Learning
• Data Science

Background:

• Software Defect Prediction (SDP) is crucial for identifying early software faults but is challenged by high dimensionality.
• Existing meta-heuristic algorithms for feature selection (FS) in SDP models suffer from high costs, local optima, and slow convergence.
• This study introduces an innovative FS technique, FSCOA, based on the Chernobyl Disaster Optimizer (CDO).

Purpose of the Study:

• To develop an advanced feature selection technique (FSCOA) for Software Defect Prediction.
• To enhance prediction model accuracy by identifying optimal features and minimizing errors.
• To overcome the limitations of existing meta-heuristic feature selection methods.

Main Methods:

• The proposed FSCOA technique was applied to twelve NASA software datasets from the PROMISE archive.
• FSCOA was evaluated using Decision Tree, K-nearest neighbor, Naive Bayes, and Quantitative Discriminant Analysis classifiers.
• The performance of FSCOA was compared against existing FS techniques (FSDE, FSPSO, FSACO, FSGA) and validated using Friedman and Holm statistical tests.

Main Results:

• FSCOA achieved superior accuracy in most instances, outperforming other feature selection approaches.
• The Friedman test ranked FSCOA highly, with an average rank of 1.75 among the studied methods.
• Holm's test indicated significant performance improvements, with p-values generally below the significance threshold.

Conclusions:

• The FSCOA procedure demonstrates clear superiority over existing feature selection techniques for Software Defect Prediction.
• FSCOA offers enhanced accuracy, effectively handles complex datasets, avoids local optima, and exhibits faster convergence.
• These advantages position FSCOA as a robust solution for overcoming challenges in current feature selection methods for SDP.