At first the thesis introduces a new concept: coverage-based pattern mining. Then based on this concept, two ensemble pruning algorithms, CPM-EP and PMEP are put forward. By the algorithms we get candidate sub-patterns,and applies coverage pattern mining and a greedy strategy to acquire candidate target ensemble. Experiments show that pattern mining is an efficient selective integration strategy, able to obtain predictive ability, small-scale target integrated classifier, the paper presents that the CPM-EP and PMEP algorithm is able to meet the demand for online learning.