| One potential application of Data Mining is the use of the discovered patterns as rules in a temporal active database (TADB). Whenever a new sequence of events occurs whose characteristics potentially match a rule, the TADB will automatically trigger some actions. This integration of the mining mechanism within a TADB raises the primary issue of "timeliness". A TADB usually puts emphasis on timing as a major condition for its actions. An action should take place within a short time limit after the occurrence of some events. On the other hand, the mining mechanism should provide the TADB with rules that are up-to-date. These two requirements cannot be easily satisfied simultaneously since, in many cases, Data Mining has to deal with databases of enormous sizes.; This dissertation discusses methods to speed up the data mining process in a non-uniform environment, in which data has different characteristics at different parts of a given database. To deal with this environment, this dissertation adapts tbe Divide and Conquer strategy to meet the requirements of data mining. To this end, this strategy divides a very large and non-uniform database into smaller and more manageable component parts. These component parts are mined separately and the results are then combined. Furthermore, the Divide and Conquer strategy can be adapted as an incremental data mining technique. Experiments are conducted to evaluate the effectiveness of these methods under various situations. They show that, in a non-uniform environment, the Divide and Conquer strategy and the incremental data mining technique are effective in reducing mining time.; This dissertation also studies branching patterns. Most current mining algorithms deal only with linear sequential patterns, whereas, in real life situations, branching patterns are also possible. This dissertation revises the current algorithms to deal with this special case. |
