Survey on Complex Event Processing and Predictive Analytics
Lajos Jenő Fülöp,
Gabriella Tóth, Róbert Rácz, János Pánczél,
Tamás Gergely, Árpád
Beszédes and Lóránt Farkas
Observing failures and other – desired or undesired –
behavior patterns in large scale software systems of specific domains
(telecommunication systems, information systems, online web
applications, etc.) is difficult. Very often, it is only possible by
examining the runtime behavior of these systems through operational
logs or traces. However, these systems can generate data in order of
gigabytes every day, which makes a challenge to process in the course
of predicting upcoming critical problems or identifying relevant
behavior patterns. We can say that there is a gap between the amount of
information we have and the amount of information we need to make a
decision. Low level data has to be processed, correlated and
synthesized in order to create high level, decision helping data. The
actual value of this high level data lays in its availability at the
time of decision making (e.g., do we face a virus attack?). In other
words high level data has to be available real-time or near real-time.
The research area of event processing deals with processing such data
that are viewed as events and with making alerts to the administrators
(users) of the systems about relevant behavior patterns based on the
rules that are determined in advance. The rules or patterns describe
the typical circumstances of the events which have been experienced by
the administrators. Normally, these experts improve their observation
capabilities over time as they experience more and more critical events
and the circumstances preceding them. However, there is a way to aid
this manual process by applying the results from a related (and from
many aspects, overlapping) research area, predictive analytics, and
thus improving the effectiveness of event processing. Predictive
analytics deals with the prediction of future events based on
previously observed historical data by applying sophisticated methods
like machine learning, the historical data is often collected and
transformed by using techniques similar to the ones of event
processing, e.g., filtering, correlating the data, and so on.
In this paper, we are going to examine both research areas and offer a
survey on terminology, research achievements, existing solutions, and
open issues. We discuss the applicability of the research areas to the
telecommunication domain. We primarily base our survey on articles
published in international conferences and journals, but we consider
other sources of information as well, like technical reports, tools or
web-logs.
Keywords: Survey, Complex Event Processing, Predictive Analytics.
Back