Applying Sequential Correlation to Identify Order Dependent Activity Pairs

Abstract

Mining sequential patterns has been one of the most useful fields in data mining. For example, one recent application of sequential pattern mining is analyzing logs of activity to predict and recognize activities. This is used for both daily life activities and virtual activities.  However, only patterns are not always enough to represent the truly interesting information to the end user, especially when the support threshold is low and the number of frequent patterns is huge. A correlation measure is necessary to decipher the relationship between the logged activities. This data is generally collected as a sequence and there is no widely popular correlation measure for elements in sequential patterns. Therefore, we define Sequential Correlation, a novel correlation measure, for discovering important knowledge in sequential patterns and the corresponding full method, SCMine, to classify patterns based on the measure. We employed the measure to establish either a unidirectional or bidirectional association among the activities within a sequence and subsequently classified the sequences based on order dependency. Moreover, an efficient implementation approach for our measure is also discussed. Our performance study shows that, a significant number of activity patterns can be pruned when degree of order among the activities is important. So, it is also useful for classifying or pruning less significant activity patterns from a vast number of frequent sequential patterns.

DUJASE Vol. 8 (1) 51-61, 2023 (January)