A Clustering Algorithm for Mixed Data Based on Lattice Theory

Hybrid data clustering analysis is an important issue in data mining. Compared with the clustering Algorithm for pure numerical data, the methods can process the data together with numerical and categorical value are quite few although the hybrid data exist in a lot of fields. After analyzing the traditional clustering algorithms, the paper presents a new algorithm to cluster the hybrid data based on lattice. The method changes the object´s attributes to lattice based on the conception of simple tuples and hyper tuples in lattice, use the numbers of covers to measure the similarity between labels, and choose the clustering mean-point according to the rule of high covers to high similarity. Experiments show that the new algorithm is more efficiently than the other classical ones.