Evolutionary algorithms have shown their powerful capabilities in different machine learning problems including clustering which is a growing area of research nowadays. In this paper, we propose an efficient clustering technique based on the evolution behavior of genetic algorithm and an advanced variant of nearest neighbor search technique based on assignment and election mechanisms. The goal of the proposed algorithm is to improve the quality of clustering results by finding a solution that maximizes the separation between different clusters and maximizes the cohesion between data points in the same cluster. Our proposed algorithm which we refer to as “EvoNP” is tested with 15 well-known data sets using 5 well-known external evaluation measures and is compared with 7 well-regarded clustering algorithms . The experiments are conducted in two phases: evaluation of the best fitness function for the algorithm and evaluation of the algorithm against other clustering algorithms. The results show that the proposed algorithm works well with silhouette coefficient fitness function and outperforms the other algorithms for the majority of the data sets. The source code of EvoNP is available at http://evo-ml.com/evonp/.

进化算法已经在包括聚类在内的各种机器学习问题中显示出了强大的功能，而聚类是当今研究的一个增长领域。在本文中，我们提出了一种基于遗传算法进化行为的有效聚类技术，以及一种基于分配和选举机制的最近邻搜索技术的高级变体。所提出算法的目标是通过找到一种解决方案来提高聚类结果的质量，该解决方案可以使不同聚类之间的距离最大化，并使同一聚类中数据点之间的内聚性最大化。我们提出的算法（我们称为“ EvoNP”）使用15种知名数据集进行了测试，使用了5种著名的外部评估方法，并与7种广受好评的聚类算法进行了比较。实验分两个阶段进行：对算法的最佳适应度函数进行评估，并针对其他聚类算法对算法进行评估。结果表明，该算法在轮廓系数适应度函数方面效果良好，并且在大多数数据集上均优于其他算法。EvoNP的源代码可从http://evo-ml.com/evonp/获得。