Data Mining Front Side

1
Data Mining Front Side
Which molecules inhibit cancer?
Data
ACME DATA-MINO-MATIC
Segmen-tation
Regression
Cleaning
Evalua- tion
Classifi-cation
Potential drug molecules
Modified version of cartoon from Union of
Concerned Scientists
2
Data Mining Backside

• Pay no attention to the man behind the
  curtain.
• -Wizard of Oz

3
Strategies
?
?
Data
Data
ACME DATA-MINO-MATIC
ACME DATA-MINO-MATIC

• Input/Output funnels are largely art
• Capture and exploit meaning and context not just
  data semantic web
• Adapt goal centered versus algorithm centered
  approach

4
Strategies
?
Data
ACME DATA-MINO-MATIC

• Sub-boxes are scientific but narrow
• Push more functionality in each box
• Grow the theory
• Move boxes closer to real world
• heterogeneous data, scalability, simplicity,
  sparseness, interpretability, interestingness

5
Strategies
?
?
Data
Data
ACME DATA-MINO-MATIC
ACME DATA-MINO-MATIC

• Funnels and levers not always published
• Mundane details matter
• Mine the mining
• Identify best practices, problem strategies, and
  emerging methods via data mining applications
  website
• Social tagging and ranking