Sequential Pattern Mining with Inaccurate Event in Temporal Sequence

In true-life, the existence of many events which are occurred in the interval may cause uncertainty in events ordering. The inaccurate event has been introduced for sequential pattern mining to improve accuracy of computing support threshold. In this paper, we store a sequence in the chain table. Sequence with inaccurate event can be expressed expediently. Besides, precise support is introduced to evaluate the probability of a pattern contained in a sequence. And the probabilistic ordering is employed to handle overlapping events. In essence, probabilities of generated candidate pattern contained in an inaccurate sequence are computed. The maximal value in probabilities is selected as precise support of the sequence. The sum of all inaccurate sequence precise supports in the database is computed. If the ratio value between the sum and the length of database is not less than predefined minimum threshold, the generated candidate pattern is called frequent pattern. So some infrequent patterns might be turned to be frequent and some interesting patterns could not be missed. Performance analysis shows the accuracy of discovering frequent patterns is improved.