一种改进的期望最大化算法

An Improved Expectation-Maximum Algorithm

针对使用传统期望最大化算法进行参数估计的混合高斯模型的最终聚类效果过于依赖初始概率密度中心的问题,提出一种基于模糊C均值算法进行参数初始化的改进期望最大化算法.实验结果表明,在实际的用户知识水平聚类任务中,与传统无监督聚类算法(模糊C均值算法、K-means算法和未改进的期望最大化算法)相比,改进的期望最大化算法的聚类性能度量指标均更优,相比于传统聚类算法全局聚类效果更好.

Aiming at the problem that the hybrid Gaussian model using the traditional expectation-maximum algorithm for parameter estimation was too dependent on the initial probability density center for the final clustering effect, we proposed an improved expectation-maximum algorithm based on the fuzzy C-means algorithm for parameter initialization. The experimental results show that compared with the traditional unsupervised clustering algorithms(the fuzzy C-means algorithm, the K-means algorithm and the unmodified expectation-maximum algorithm), the improved expectation-maximum algorithm has better clustering performance metrice and better global clustering effect than the traditional clustering algorithm in a practical user knowledge level clustering task.