Generalization of set of features for the software fault prediction using ant colony-based feature selection

Abstract The inability to clearly define the necessary features and the number of features needed for effective software fault prediction has been one of the main problems. Current feature selection methods often fail to reliably identify the optimal feature subset. Predictive outcomes become unpredictable as a result. We suggest a generalized feature selection technique that makes use of the ant colony optimization (ACO) algorithm in order to solve this problem. To guarantee accurate feature selection, this algorithm is run several times. The effectiveness of the chosen features is demonstrated by our evaluation using a dynamic classifier. Experiments on a unified dataset demonstrate that our approach achieves 94.12% accuracy and effectively reduces dimensionality. This offers a reliable way to predict faults in a variety of datasets.