A General Feature Abstraction Method for Clustering Algorithm

Clustering algorithms are one of the most important methods in data mining, which can be used in real applications to detect the similarity of objects. To improve the performance, we use a general feature abstraction method, the Auto-Encoder[9], which can compress the data features so that the noises of data can be filtered. Our extensive experimental results show that when we use a more complicated neural networks with certain activation functions, we can significantly improve the performance, both the computing cost and the clustering accuracy, of the clustering methods.