The feature selection problem is known as one of the most critical issues with several applications in classification. Certain features of a dataset do not usually contain useful information or play important roles in the data classification. It is possible to reduce the computational burden through the elimination of unnecessary features. This paper was conducted to propose an opposition-based binary competitive optimization algorithm, namely OBCOOA, to solve the wrapper-based feature selection problems. The basic idea of the competitive optimization algorithm (COOA) lies on the natural competition between different species to survive, which is a continuous algorithm inherently. The present study entailed the main contributions as follows: primarily, a time-varying V-shape transfer function was used in the optimization process to present the binary version of the COOA algorithm, which established the right balance between the exploration and exploitation phases; secondly, in the proposed algorithm, the opposition-based learning mechanism was utilized to improve the diversity quality within the population members and to incorporate a suitable initial population. In the feature selection problem, the classification error rate and the number of selected features objective functions are usually adopted, which causes this problem to convert into a multi-objective optimization one. This research presented a single- and multi-objective approach from the proposed algorithm in order to solve the feature selection problem. The proposed algorithm was applied to 27 benchmark datasets, and the evaluation results were compared to the other well-known binary optimization algorithms. The experimental results indicated that the proposed algorithm has a better ability to find an optimal subset of features with the least classification error rate and the number of the selected features.

特征选择问题被认为是分类中多种应用中最关键的问题之一。数据集的某些特征通常不包含有用信息或在数据分类中起重要作用。可以通过消除不必要的特征来减少计算负担。本文旨在提出一种基于对立的二元竞争优化算法，即 OBCOOA，以解决基于包装器的特征选择问题。竞争优化算法（COOA）的基本思想在于不同物种之间的自然竞争才能生存，它本质上是一种连续算法。本研究的主要贡献如下：在优化过程中使用了随时间变化的 V 形传递函数来呈现 COOA 算法的二进制版本，从而在探索和开发阶段之间建立了适当的平衡；其次，在所提出的算法中，基于对立的学习机制被用来提高种群成员内的多样性质量，并合并一个合适的初始种群。在特征选择问题中，通常采用分类错误率和选择特征目标函数的数量，这使得该问题转化为多目标优化问题。本研究从所提出的算法中提出了一种单目标和多目标方法，以解决特征选择问题。所提出的算法应用于 27 个基准数据集，并将评估结果与其他著名的二元优化算法进行了比较。实验结果表明，所提出的算法能够更好地找到具有最小分类错误率和所选特征数量的最优特征子集。