Biomedical Engineering Reference
In-Depth Information
On the Horizon
Real-time data mining, sometimes referred to as transaction monitoring, is rapidly gaining in
popularity because of its increased value over the traditional mining of a data warehouse or
database. In many industries, just-in-time analysis of data is much more valuable than analysis of
data dredged up from the past—even if the past is only an hour or two removed from the present.
For example, transaction monitoring is used by the credit card industry to detect fraud. As soon as a
questionable transaction—a major purchase from a vendor not frequented by the legitimate card
holder, for example—is detected, the system flags the point-of-sale system and the retailer has to
call the credit card company for authorization. Mining the data even 30 seconds after the transaction
is complete is of relatively little value, especially if the thief disposes of the card after the purchase.
In clinical medicine, real-time data mining is being used to detect potential drug-drug interactions,
allergic reactions, and other side effects at the time a prescription is ordered. In some instances, a
drug already in the body can potentiate another drug, causing the patient to overdose on the second
drug, even though the dosage would be therapeutic without the other drug in the body. An overdose
may result, for example, because both drugs are eliminated by the same pathway in the liver, and
one drug completely saturates the metabolic pathway for drug elimination. Obviously, data on
possible interactions is pertinent only before a patient is accidentally given the wrong drug or wrong
dose of the appropriate drug.
The same technology can be extended to provide real-time analysis of drugs against a patient's
genome, enabling the just-in-time delivery of custom drugs or as a means of detecting likely side
effects of standard drugs on a given patient. In the bioinformatics laboratory, real-time data mining
of results as they are generated by a sequencing machine or microarray reader can provide
researchers with indicators as to the value of the data, error rate, and relatedness of the data to
previous studies.
Three technologies that support real-time data mining are real-time capture, message-oriented
middleware, and rule-based systems. Real-time data capture intercepts data from the source, before
it is written to the database or data warehouse. This allows comparison of data to be made without a
time-consuming data extraction process. Similarly, message-oriented middleware captures
transactions, takes them off-line in batches, and stores the data in high-speed RAM (see Figure 7-
18 ). While in RAM, the data are mined using a high-performance database manager with powerful
RAM-based data handling features. The third technology, rule-based systems, can be used to create
filters that intercept only those real-time transactions fitting a profile defined in easily edited rules.
The data selected by the filter can then be rapidly mined using conventional processors or RAM-based
technologies, as dictated by the performance limitations of the system.
Figure 7-18. Real-Time Data Mining.
 
 
Search WWH ::




Custom Search