AIxploratorium
Decision Trees - Page 9

Decision Trees -- Conclusions

Synopsis

Congratulations! You now know essentially all of the important points related to learning decision trees... as well as many points seminal to learning in general:

• why learning a classifier can be useful
• what a decision tree is
• how to write a basic decision tree learner
• the role of information theory
• overfitting and pruning

Of course, we have provided just the tip of the iceberg; there are many many other areas to investigate --- some fairly well resolved, while others remain topics of very active research, which will continue to lead to interesting, and financially rewarding, results.

We list some of them below.

By the way, you'll be pleased to know the MallRats did win the game.

Other Topics/Issues

• Business Successes
  Does anyone actually using this technology?
• Modifications to the LearnDT algorithm
  Why uses these particular splitting functions? ... pruning functions? Etc.
  What if the attribute values are continuous?
  Etc.
   
• Other Learning Models
  So far, we assumed our goal was just to predict which team would win.    What if, instead, you wanted to know the point spread? Or ... what if you were the MallRat coach, and wanted to know how to make our team win? Could you use this tree? Why or why not?
   
• Articles
  • "Decision Trees", by FamousDecisionTreePerson, 2000.
     
  • Learning in General
    So far, we have focused on a particular formalism for representing information -- viz., a decision tree. There are many others, including rule-based systems, neural nets, and bayesian networks. Each has its own learning algorithms. What is in common across these representations, and what is significantly different? And what about other types of learning, which are not simply for the purpose of classification.

    Also address the important issue of VALIDATION: Is the tree learned really correct?
    For that matter, what does it mean to say a tree is correct, anyway?