Summary

• KDD is motivated by large volumes of data to be analyzed and builds on DB, ML, statistics
• ILP is needed for multirelational, structural problems

 

• classification vs. description tasks
• one important description task: subgroups discovery
• algorithm MIDOS

 

• KDD systems need to support dynamic nature of KDD process
• extensibility can be realized by plug-in architecture

Kepler system

 

---