An Adaptive Rank Aggregation-Based Ensemble Multi-Filter Feature Selection Method in Software Defect Prediction.

Area of Science:

• Software Engineering
• Machine Learning
• Data Mining

Background:

• High dimensionality is a significant challenge in software defect prediction (SDP).
• Selecting optimal features using filter feature selection (FFS) methods in SDP remains an open research problem, termed the filter rank selection problem.

Purpose of the Study:

• To propose a novel adaptive rank aggregation-based ensemble multi-filter feature selection (AREMFFS) method.
• To address both high dimensionality and the filter rank selection problem in SDP.

Main Methods:

• The AREMFFS method assesses and combines the strengths of individual FFS methods.
• It aggregates multiple rank lists to generate and select top-ranked features for SDP.
• The method was evaluated using decision tree (DT) and naïve Bayes (NB) models on diverse defect datasets.

Main Results:

• AREMFFS demonstrated superiority over baseline FFS methods, existing rank aggregation methods, and its own variants.
• The proposed method significantly improved the prediction performance of SDP models.
• It effectively resolved high dimensionality and filter rank selection challenges.

Conclusions:

• Combining multiple FFS methods is recommended to leverage individual strengths and filter-filter relationships for optimal feature selection in SDP.
• The AREMFFS method offers a robust solution for enhancing software defect prediction accuracy.