back to top

Sequence Detection Algorithms have their origin in basket analysis and web mining though they have successfully been applied to fraud detection and analysis of commodity prices.

An example rule discovered by CAPRI when applied to crude oil prices in the New York and London exchanges is shown below:

If price in London increases by less than 0.9% and the Next Day there is a drop in the NYMEX WTI price of between 0.79%   and 1.55% and a drop in London price of between 0.96% and 1.68%

Then (On the Following Day)
there will be a further drop in the NYMEX WTI price of between 1.5% and 2.45%.

Confidence in the rule is 75%
This pattern appeared in 9% of all time windows being analysed.

back to top

Discovery of such rules is a departure from traditional sequence detection. Traditional sequence detection generally dealt with discrete events occurring in time as opposed to a continuous series as is the case with commodity prices.

Capri belongs to the Apriori-family of data mining algorithms, with its origin in association rule discovery. Capri is used within the third party applications to discover different types of sequences across records (and therefore, over time). Prior to Capri, general sequence questions could not be answered unless you knew the sequences in advance or had narrow constraints on the problem. Typical sequential patterns that can be found in data sets using Capri are:

• Products bought by customers across multiple transactions.
• Financial transactions made by a business in a fiscal year.
• Clickstreams or Web site paths for understanding purchases, exits, traffic, and crime on the Web.
• Frequent sequences of the chemical bases that make up human DNA.
• Patterns of non-compliance or fraud over time.

back to top

Association Algorithms Association Algorithms such as GRI and Apriori (the algorithm that Capri is based on) generate rules showing which things (events, attributes, purchases, etc.) typically occur together. Using an association algorithm one produces a list of rules. The rules describe the conditions under which certain conclusions occur. A typical rule from GRI/Apriori might look like this:

conclusion <= conditionA & conditionB & ...
beer <= snacks & newspaper

This rule is interpreted as follows: Customers who buy snacks and a newspaper are also likely to buy beer.

Note: This rule does not show a causal relationship; it is merely showing the likelihood of certain things occurring together. Association rules normally include information on:

• Coverage (or Support). Indicates how often the conditions and conclusion occur together.

• Accuracy (or Confidence). Indicates how often, when the conditions occur, that the conclusion also occurs

back to top

back to products