Questions to think about

  1. Do I really understand how my technique works? Are there important variations? How computationally expensive is it?

  2. What kinds of data mining problems is my technique especially useful for? What kind does it not work well for?

  3. How does my technique relate to other techniques? Can my technique be a useful preprocessing step? A useful postprocessing step?

  4. How stable is my technique? If the training data changes, do I get a completely different model/pattern? Do I get a model/pattern that's different internally but implies the same conclusions?

  5. Which techniques seem to have been most productive on this dataset? Which ones seem to have been least productive? Can you identify properties of the dataset that might explain this, or do you think your conclusions are more general?

  6. What would I have done differently if I were to start modelling this dataset starting from a clean slate?