Abstract:
PrefixSpan is a sequential pattern mining algorithm used to analyze a set of
event transactions and discover relationships among them. Its main idea is to examine
only the prefix sub-sequences with minimum support. PrefixSpan is efficient in terms
of memory requirement and response time. However, it comes up with a large
number of sequential patterns, only a few of which are interesting.
This thesis describes the underlying mechanism which augments the
PrefixSpan algorithm. Specifically, a new mechanism which additionally considers
minimum confidence of patterns to automatically filter out uninteresting and useless
discovered patterns was developed. In addition, the mechanism includes data preprocessing
and data post-processing such as data cleaning, verification, reformatting,
and time duration calculation. The mechanism was evaluated on a public phone
malfunction dataset.
The experimental results showed that the augmentation of PrefixSpan can
reduce the number of uninteresting patterns and increase the discovery capacity. This
research contributes to public phone maintenance planning whose goal is to minimize
the number of malfunctions, the transportation expenses, the time and the manpower
for maintenance.