Quantitative estimation of time interval of 3-sequences

Constraint-based sequential pattern mining algorithms discover sequential patterns among from sequence data and the resultant sequential patterns satisfy a given constraint. For time stamped sequences duration and/or gap constraints can be applied to obtain corresponding constraint-based sequential patterns. One of the shortcomings of existing algorithms is the requirement to pre-specify a time window to generate time constraint-based sequences. Another limitation is that although these sequences can predict about events that would follow each other, intermediate time interval of these sequences is not available. To overcome these issues, we propose to focus on the estimation of an average intermediate time interval between events that follow each other as a sequence. Instead of sliding a pre-specified time window over sequence data, for a combination of any three events we determine how often these events follow each other as a sequence. A minimum support threshold `min_sup' is used to verify if any given 3-sequence is frequent or not. Earlier, we have proposed an algorithm to enlist inter-transaction associations using 2-sequences [22]. We have extended this work to obtain a list of frequent 3-sequences. We obtained time intervals between successive events of sequences and furthermore we have obtained sequences that satisfy range interval constraints too. The results of our experiments on live retail shop data set are presented.