Research Challenges for Data Mining (2) Automation (make it easy for end-user to deploy or use)

• Wizards to help:
• define the mining task, inputs, outputs; modeling noise and missing data
• fusing prior knowledge to help in model search and fitting (Bayes Nets)
• Data transformation and dimensionality reduction, coping with curse of dimensionality
• Guarding against overfit, chance fitting, and other traps
• Modes of growth of data: “how does the data grow?”
• Interface to derive simplified forms of the extracted models for comprehensibility and visualization (trade-off simplicity and accuracy for sake of understandability)

---