EVENT PROCESSING IN SENSOR STREAMS - Managing and Mining Sensor Data - page 93

4.1 Probability of Events
Uncertainty of events is among the most important challenges of com-
plex event detection, and there can be various reasons that produce prob-
abilistic event data. Typical cases include conflicting readings, missed
readings, or granularity mismatch.
Complex event detection on probabilistic data can be divided into two
categories: local uncertainty detection and global uncertainty detection.
When an tuple or object is independent of others, and the event de-
tection only concerns with the uncertainty of itself, it is called local
uncertainty detection . On the other hand, when the event detec-
tion must consider the combined uncertainty among objects, it is called
global uncertainty detection . Generally, if the decision on whether
an object satisfies a detection condition depends on other objects not
involved in the same generation rule, global uncertainty has to be con-
sidered. Semantically, we have to examine the possible worlds one by
one and count the probability that a combination of objects or tuples is
an answer.
Probabilistic event processing has been studied in the context of query
processing over probabilistic data streams. Jayram et al. [48] introduce
a probabilistic stream model. Jayram et al. [48, 49] and Garofalakis [50]
propose ecient algorithms for computing aggregate functions over un-
certain data streams, where correlations across time are not considered.
A hidden Markov model is used in [51] to support queries over probabilis-
tic streams produced. The queries are limited to selections, projections,
and aggregations. The method proposed in the Data Furnace project
[53] extracts probabilistic events from imprecise sensor data. Its design
relies on exploiting an inference engine to compute event probabilities,
for example, using the work in [52].
Lahar[52] is an event processing system for probabilistic event streams.
Lahar supports a much richer query model over probabilistic streams in-
cluding sequences and joins. By exploiting the probabilistic nature of the
data, Lahar yields a much higher recall and precision than deterministic
techniques operating over only the most probable tuples.
4.2 Disorder of Events
The tuples in an event flow may be ordered or disordered on some
attributes. When an order exists, some operations become easier and
can be performed without the need of arbitrary storage; however, when
this order is violated, it will be called “ event disorder ”. Poset pro-
cessing consists of performing operations on a set of tuples that may not
be related by a total ordering. Any partially ordered set of tuples can