Automatic parameter determination- Automatically working out the most appropriate values for parameters that would otherwise have to be supplied.
User Interfaces- Input interfaces
- Process Monitors
- To see the algorithm working (even if just a crude process bar).
- Rule Visualisation
- Very, very useful is sense it to be made of many kinds of rule.
Knowledge Bases- Useful to help discriminate between known and unknown rules.
Interactive Mining- Iteratively (ie. process runs to completion many times)
- Guided (ie. process is guided as it runs).
Incremental Algorithms
Behaviour Analysis. - Analysing actors spending habits, categorising customers likely to default, vote differently, be committing fraud, players getting tired in the last quarter of a match, etc.
Cross Selling. - Selling more of a range of products to those customers who already have a given product from a given organisation. Examples in this category include book sellers, insurance and financial institutions, but might also include political parties and university alumni associations in more indirect ways.
Direct Mailing. - Direct mailing aims to target promotional mail to potential customers who fit some profile. Given that the response from e-mail (which is a cheap delivery mechanism) is very low, higher cost mailing methods are often used and thus eliminating unlikely purchasers may be important.
Churn Modelling. - Churning is activity that leads to no additional benefit for an organisation. For example, in order to make a course more attractive, a University might develop a new strand to that course. If the result is simply the transfer of existing students from other courses then the effect is said to be churning. Similar effects can be seen in new offerings from finance companies, new flights by airlines, etc.
Market Basket Analysis. - The analysis of items grouped together on some identifier from which co-occurring items can be found.
Personalisation. - Improving either presentation (ie look-and-feel) and/or product mix offered based on individual tastes.
Prediction. - Improving on human capabilities by taking into account more variables, recent changes to behaviour, etc.
Note that many of these represent future directions for data mining research. Note that many of these represent future directions for data mining research.Health and Medical. - An area in which properly controlled use of data mining (ie. data mining under the guidance of a robust knowledge discovery process) holds the promise of substantial benefits. To date, relatively few examples of full scale data mining have been reported.
Defence Industry. - In many ways, many of the technologies used in medical informatics are similar to those used in the defence industry (although of course the application is different). For example, the technology used in constructing the complex models to predict the spread of disease in the community through tracking carriers is similar to that used to track battlefield events. Techniques for analysing incomplete or inaccurate information is similarly congruent.
DNA Sequencing and Biomedical Applications. DNA Sequencing and Biomedical Applications. - DNA sequencing attempts to search the 100,000 genes in the human DNA of known cases of people with genetic disease (or disposition to a disease) for commonalities in order to discover which genes are responsible for the disease and thus provide either a screening process or hints towards a cure.
Web Site Performance. - Many paths through a series of web pages are common. The performance of web sites can thus sometimes be improved by pre-fetching the most commonly accessed subsequent page. Web sites might also be personalised based on previous activity.
Telecommunications. - The telecommunications industry uses data mining in a variety of application domains including quality of service applications, bottleneck analysis and fraud.
Fraud Detection. - This was one of the first major applications of data mining and has included:
- Profiling the sorts of commands entered by hackers - in Unix, they might be commands such as ls, cat, more, less, cd, etc. but would exclude editing, creating directories, starting applications, etc. A daemon might monitor all activity and report as suspicious users with a hacker profile.
- Spotting outliers. This has been done, for example, in Australian Tax Office auditing, medical benefits checking (to stop so called Doctor shopping, medical practitioner claims (to prevent overbilling).
Sports. - This involves the complex use of multimedia technology to allow coaches to detect problem behaviour and has been implemented in, at least, American Rules Football and Basketball. A smaller system with some of the required aspects (but without explicit support for data mining) was also implemented in Australian Rules Football in the late 1980's. Also included here might be result prediction strategies (particularly in league sports and horse racing).
Dostları ilə paylaş: |
