Ensemble of semi-supervised feature selection algorithms to reinforce heuristic function in ant colony optimization

Feature selection (FS) is a well-known dimensionality reduction method that chooses a hopeful subset of the original feature collection to diminish the influence the curse of dimensionality phenomenon. FS improves learning performance by removing irrelevant and redundant features. The significance of semi-supervised learning becomes obvious when labeled instances are not always accessible; however, labeling such data may be costly or time-consuming. Many of the samples in semi-supervised learning are unlabeled. Semi-supervised FS techniques overcome this problem, simultaneously utilizing information from labeled and unlabeled data. This article presents a new semi-supervised FS method called ESACO. ESACO uses a combination of ACO algorithm and a set of heuristics to select the best features. Ant colony optimization algorithm (ACO) is a metaheuristic method for solving optimization problems. Heuristic selection is a significant part of the ACO algorithm that can influence the movements of ants. Utilizing numerous heuristics rather than a single one can improve the performance of the ACO algorithm. However, using multiple heuristics investigates other aspects to attain optimal and better solutions in ACO and provides us with more information. Thus, in the ESACO, we have utilized the ensemble of heuristic functions by integrating them into Multi-Criteria Decision-Making (MCDM) procedure. So far, the utilization of multiple heuristics in ACO has not been studied in semi-supervised FS. We have compared the performance of the ESACO using the KNN classifier with variant experiments with eight semi-supervised FS techniques and 15 datasets. Considering the obtained results, the efficiency of the presented method is significantly better than the competing methods. The article's code link on GitHub can also be found at the following: https://github.com/frshkara/ESACO.