In this paper, a new strategy for classification based on sequential association rules is presented. It removes potential problems from a poor selection of training dataset, keeping a high accuracy with a smaller number of rules for classes. Furthermore, it considers all rules in a class with the same power decision to classify any transaction. In the experiments, the proposed strategy is compared against sequence classification approaches as SVM, J48, and others classifiers. The results show that the proposal has better accuracy than already reported classifiers.