Set up a new experiment and select the \breast-w.arff" and \primary-tumor.arff datasets. Add the algorithms ZeroR, and J48 with default parameters. Also add four more instances of J48 where the \Con dence" (C) is set to 0:1, 0:01, 0:001 and 0:0001, respectively, keeping all other parameters set to their default values.
Tree pruning Compare the C parameter settings by: (i) \Percent correct" using a baseline of ZeroR; (ii) \Percent correct" using a baseline of J48 with default parameters, and (iii) \measureTreeSize" using a baseline of J48 with default parameters. Save these results to a new le called \q2.out".
Results interpretation Answer this in the le \answers.txt".
2a) Looking at the predictive accuracy (Percent correct) results for tree learning on these data sets, has learning improved accuracy over the baseline ?
2b) Looking at the tree sizes on these data sets, how do these change with variation in the con dence values ? Does this a ect accuracy ?
2c) What do you observe about the change in accuracy, with respect to the baseline of ZeroR, on both datasets ? Is there a di erence in this measure between datasets ? If so, how do you explain this in terms of the datasets?