Simple Event Detection

All the time sequences from the sample domain, computer system performance, are numeric. In order to demonstrate Apriori Sets And Sequences events must be detected in these numeric time sequences. The events used are Increase, Decrease, and Sustain. These are very simple. They are simple so the results can be easily understood. They can be used in conjunction to describe any numeric time sequence.

It is important to note that the events detected should have some meaning to the user and to the domain from which the data set comes. These events will be used in the association rules produced. If their meaning is not clear to the user the rule will not benefit the user.

Increase event

is defined as a sequence of values where each subsequent value is larger than the one before it.

Decrease event

is defined as a sequence of values where each subsequent value is smaller than the one before it.

Sustain event

is defined as a sequence of values where each subsequent value does not differ from the mean of the values by a user set tolerance.

These are a very strict definitions but sufficient for demonstrating Apriori Sets And Sequences in this domain.

The event detection was implemented as a Weka filter. Filters are the main means for preprocessing data in Weka. The event filter reads in a data set containing time sequence attributes. For each time sequence attribute and each type of event pair, it adds an event attribute if that event type is detected in the time sequence. The individual events' begin and end time is noted according to the real time line for a particular instance.