Database Reference
In-Depth Information
relational knowledge discovery
system uses CBR to identify the transactions by
the alert filter and process the features of current
situation, such as the interest rates, company
news and so on.
However, CBR is an approach that accepts
anecdotal evidence as its main operating principle.
There is no guarantee that the solutions for the
input are correct without the statistically relevant
data to test and to support it.
The National Association of Securities Dealers
(NASD) (Bessembinder 1999) uses the probability
relational model to identify the most dangerous
brokers' trade. The NASD is a self-regulatory or-
ganization of the securities industry responsible for
the operation and regulation of the Nasdaq stock
market and over-the-counter markets. It watches
over the Nasdaq to make sure that the market
operates correctly. In 2007, the NASD merged
with the New York Stock Exchange's regulation
committee to form the Financial Industry Regula-
tory Authority, or FINRA.
Neville et al. (2005) reports the usage of re-
lational probability trees (RPTs) for the task of
surveillance. RPTs extend probability estimation
trees to a relational setting. Due to their selectiv-
ity and intuitive representation of knowledge,
tree models are often easily interpretable. This
makes RPTs an attractive modeling approach for
NASD examiners. The RPT learning algorithm
adjusts for biases towards particular features due
to the unique characteristics of relational data.
Specifically, three characteristics, concentrated
linkage, degree disparity and relational autocor-
relation, can complicate the efforts to construct
good statistical models, leading to feature selection
bias and discovery of spurious correlations. By
adjusting for these biases, the RPT algorithm is
able to learn relatively compact and parsimonious
tree models.
It is successful in reality, but it is difficult to
run the above system for other stock exchanges in
that it depends heavily on the qualified data col-
lected by Central Registration Depository (CRD).
NASD's task of ranking brokers for examination
has three characteristics that are common to many
knowledge discovery tasks, but that are rarely
addressed in combination. Accurate ranking of
brokers is inherently probabilistic, relational,
and temporal.
expert System (eS)
An expert system (Lucas 1993) is a traditional
application of artificial intelligence which aims
to reproduce the performance of one or more hu-
man experts. Most expert systems are commonly
used in a specific problem domain. Expert system
is able to analyze information input by using a
set of rules. The expert system may also provide
mathematical analysis of the problems, and has
the ability of reasoning to generate solutions. A
simple decision tree or fuzzy logic may be utilized
in an expert system. An important component of
expert system is the aid of human workers or a
supplement of information.
The New York Stock Exchange (NYSE) uses
a series of ES called ICAS (Integrated Computer-
Assisted Surveillance) (Francis 1989). The ICAS
system has the abilities of identifying questionable
trades and analyzing the trades with other external
information. It finally specifies the most likely
instances of insider trading and produces alerts
for supervisors.
Lucas (1993) reports on the development of
an experts system for stock market surveillance
in the America Stock Exchange (AMEX). The
system provides recommendations and a number
of significant data for regulators. For example, it
computes and displays the maximum potential gain
or loss avoidance from insider trading based on
the information of price, volume, etc. The expert
system finally prints a report with salient data and
related information for audit trail purposes.
Search WWH ::
Custom Search