I. S¸irin, "Learning with Feature Partitions" M.Sc. Thesis, Bilkent University, Dept. of Computer Engineering and Info. Sci., Tech. Rep. No. BU-CEIS-9312.  Home/Search   Document Details and Download   Summary   Related Articles

....one of the nearest partitions to the left and the right of the new example is in D f (generalization limit) distance and of the same class as the example, then it is generalized to cover the new feature value. Otherwise, a new point partition that corresponds to the new feature value, is inserted [7]. A version of CFP called GACFP has been implemented to learn these parameters of the CFP using a genetic algorithm [4] No similarity and distance metric is used for prediction in CFP. Prediction process is performed according to local knowledge of each feature. The classification process of the ....

....value of the old partition to the newly formed partitions. If the representativeness values of any of the resulting subpartitions drop below the confidence threshold times the observed frequency of its class, then that subpartition is removed from partition list of the feature; see [7] for details. Depending on the noise level of the data set and the number of the irrelevant attributes, the value of the confidence threshold changes between 0 (do not remove any partition) and 1 (remove a partition if its representativeness value drops below the observed frequency of the its ....

I. S¸irin, "Learning with Feature Partitions" M.Sc. Thesis, Bilkent University, Dept. of Computer Engineering and Info. Sci., Tech. Rep. No. BU-CEIS-9312.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC