RULES - TL: A SIMPLE AND IMPROVED RULES ALGORITHM FOR INCOMPLETE AND LARGE DATA
Hebah ElGibreen · 2013
In every aspect of life, many information is flowing around in which human can take advantage of. Thus, machine learning was born to gather such information and learn how it can autonomously deal with a certain problem. Different domains have grown to serve different purposes in the machine learning field. One of the mostly active domains in this field is Inductive Learning. One algorithm family developed in inductive learning is RULES. Specifically, it is a covering algorithm family where rules are directly induced from a given set of examples. However, two major deficiencies were found in the algorithms that belong to this family. They need to tradeoff between time and accuracy while searching for the best rule and incomplete data were inappropriately handled. Consequently, this paper proposes a new algorithm that is built based on RULES-6. It uses an advance machine learning method called “transfer learning ” to gather the knowledge of other agents in different domains and use it as the base knowledge that would reduce the time of search. In addition, the transferred rules are also used to fill missing classes in order to consider not only the labels available in the target task but also the possibility of future cases. Finally, the performance of the proposed algorithm will be tested and compared to other rule induction algorithms to prove that it actually improved the accuracy, reduced the error rate, and consumed a small amount of time.