Binary Enhanced White Shark Optimization Algorithm and Its Application to Semi-Supervised Feature Selection

With the development of information technology, the generation and storage of data have become convenient, resulting in a large amount of high-dimensional data. Due to the high acquisition cost of labeled data, semi-supervised learning of partially labeled data has emerged. Feature selection aims to select the most representative and informative feature subset from the original feature set and avoid the curse of dimensionality. By utilizing semi-supervised feature selection, valuable information from unlabeled data can be better extracted, thereby enhancing the model's robustness. White Shark Optimizer is a commonly used searching method, but it may fall into a local optimum when dealing with semi-supervised feature selection problems. This study proposes an Enhanced White Shark Optimization (EWSO) with a dynamic dual elite strategy and sinusoidal mutation operation. The former can make full use of the current optimal location information and improve the ability to find potential optimal solutions, and the latter can help the algorithm jump out of the local optimum and improve the accuracy of the algorithm's convergence. For dealing with discrete problems, we further introduce a binary mapping function to the EWSO, abbreviated as BEWSO. Compared with six other meta-heuristic feature selection techniques on nine high-dimensional datasets, experimental results show that the proposed method is superior to other methods. Under 80% and 60% unlabeled data, the average accuracies are 5.03% and 3.64% higher than the suboptimal value, respectively.