Analysis of the Statistical Characteristics in Mining of Frequent Sequences

Probabilistic Sequence Mining – Evaluation and Extension of ProMFS Algorithm for Real-Time Problems

“…In case of equal values, the element with less occurrences in the model sequence is chosen. After model sequence C is of length l the first step of algorithm stops and another algorithm (such as GSP) is used with model sequence C as input instead of longer main sequence S. Presented algorithm is using modified criteria from [9] that are more accurate then basic criteria presented in [2]. For simplicity presented formulas (4) and ( 5) are for case with two elements with equal values.…”

“…a) field in T corresponding to 'A' is zeroed; T = 0 2 0 ; b) distances in vector T are lowered by one; T = 0 1 0 ; c) minimal distances are not updated as M [1] [2] had the same value and zeroes are ignored as there cannot be zero distance between elements. Such value means that given combination is not present in the sequence; d) as n = 4 = L [1] this step is skipped -and from now on algorithm is not searching for 'A' signs.…”

“…a) field in T corresponding to 'B' is zeroed; T = 0 0 0 ; b) as all the elements of vector T are zeros this step is skipped; c) minimal distances are not updated as all elements in T are zeros; d) as n = 5 = L [2] this step is also skipped.…”

“…One of the algorithms used for sequential pattern mining is ProMFS algorithm (probabilistic algorithm for mining frequent sequences) proposed by Tumasonis and Dzemyda in [2]. This two-step procedure is based on the estimation of statistical characteristics of the main sequence, such as a probability of element occurrence and an average distance between different elements.…”

“…The frequency estimation of the subsequences in the sequence is based on the results of applying GSP algorithm to a model sequence. The ProMFS algorithm is meant to produce slightly less accurate results in shorter time than classical approach algorithms for very large data sets [2].…”