Data Imbalance Correction in Feature Selection for Large Scale Product Lines

In this research study, we present a pioneering feature selection approach designed specifically for product line software quality engineering to deal with the problem of too many features and imbalanced datasets. Our method combines iterative data sampling and feature ranking followed by combining the outcomes to optimize feature selection. Two conventional methods were used to evaluate this innovative technique, one involved only using filter-based feature ranking on the original dataset, while another employing just one data sampling with feature ranking. The experimental validation that is focused on different types of software product line datasets demonstrates that our method of iterative feature selection consistently outperforms traditional approaches, especially in cases where there are highly imbalanced datasets.