Abstract: Spectral clustering algorithms are dependent on the initialization of the data，the clustering results are different when input data are not identical.To solve the problem，a spectral clustering based on genetic algorithm（GASC） is proposed，which overcomes the sensitivity of the initial data and get the more stable clustering result.Compared with the improved k-means algorithm and spectral clustering，the experiments show that the suggested algorithm has better clustering performance on both artificial and UCI data.

Key words: spectral clustering, genetic algorithm, spectral graph theory, k-means algorithm, machine learning

摘要： 传统的谱聚类对初始化数据敏感，聚类结果随不同的初始输入数据而波动。针对上述问题，提出了一种基于遗传算法的谱聚类算法，该算法克服了谱聚类算法对初始数据的敏感性，得到较稳定的聚类结果。与遗传k均值和谱聚类算法相比，该算法在模拟数据和UCI数据集上获得了较好的聚类性能。

关键词: 谱聚类, 遗传算法, 谱图理论, k均值算法, 机器学习