Boosting As a Method of Novelty Detection

摘要

Interestingness of generated rules has been an active research area but novelty detection has received little attention. A pattern is novel to a person if he or she did not know it before and is not able to infer it from other known patterns. No known data mining system represents everything that a user knows, and thus, novelty cannot be measured explicitly with reference to the user's knowledge. Since in construction of decision trees with boosting each time we focus on errors and ignore the constructed tree, the idea was used to find possible novel patterns in medical domain. The first tree is more probable to present known knowledge on the field since it presents more general patterns with high coverage. Consecutive trees may present knowledge that is more novel. Thus, the main mentality behind the method is use of errors in iterative steps make the system 'delete known knowledge' and construct base of new ideas in the data. However we were interested in novel correlations with statistical significance. Therefore, based on the idea from OSDM [1], we created a model to find possible correlations. The experimental results are very promising and some of the results of the analysis are bound for submitting to medical journals.