Data Mining-Proceduralised? PAKDD-98 Panel Melbourne, April 1998

1. ACSys Hot Spots Data Mining Perspective
2. Support for the KDD Process: Standardisation
3. Issues for Standardisation

ACSys Data Mining Perspective

• Databases and Problems from HIC, NRMA, ATO, MSSSO, Medibank
• Data Mining teams contain
  • collaborator staff
    domain experts and technical data managers
  • data mining expertise
    machine learning, statistics, visualisation, database
• Real-world problems driving various back-room research groups

ACSys Data Mining Perspective

• Real-world problems driving various back-room research groups
• Data Mining GUI and Glue in Java: the Arcade Data Explorer
• Integrate Tools we use:
  • Darwin, C5.0
  • SAS, SPlus, Matlab
  • Cbos, PRIM, RPart, MARS, GAMAN, NN, GP, Evolve

Hot Spots Data Mining

How we go about Hot Spots Data Mining with the Arcade Data Explorer:

A Data Mining Architecture

Standardisation Issues

• Plug-n-Play Opportunities in Data Mining?
• Need to solve the front end and back end problems:
  • Common interface for Data Management
    • ODBC/JDBC piggy back on Warehouse/OLAP
    • Applications Oriented Databases
    • Persistent Programming Languages
  • Common Knowledge Discovery Language
    • Uniformly but not Restrictively communicate Discoveries
    • Results communicated to User and other Agents

Attempts to Proceduralise/Standardise Data Mining

• SAS and SPlus (c.f. GainSmarts)
• Applications Oriented Databases
• MLC++
• Weka
• ArcadeDX (Open/Extendible Architecture using Java, JFC/Swing, JDBC)
• CRISP/CHESS/Clementine

… to Facilitate the Creative Processes

• Data Preparation: Extracting the Right Data
• Derived Variables: The rest is simple
• Determining Interestingness:
  Collaborate with domain agents
  • Languages to express Expectations
  • Visualisation
  • Evolutionary Approaches
• The Loop to Eternity: Process Management