A New Classification Algorithm Based on Emerging Patterns

As one of the most important part in classification branches, emerging patterns has become one hot research area. Classification methods based on emerging patterns are put forward for large datasets in recent years. One of them is DeEPs and it has been proved a good classification method. This paper proposes a novel classification method-MDeEPs. On the base of DeEPs algorithm to mining EPs, allocating misclassified test instances to the correspond class for adjusting deviation. Sequentially processing each test instance and at the same time counting error number. While meeting the condition that test instances has reached a certain number and within the scope of error rate, the algorithm ends. Otherwise, perform the entire process cyclically. Each class of EPs derived from different test instance is recorded. All of them will be aggregated to build classifier. Experiments show that the improved algorithm has a good performance.