Lab Machine Learning Using Weka

Question 1

Apply both J48 and the MultilayerPerceptron to the following data sets: "bank-data.arff" and "mnist-small.arff". Note down the accuracy (5-fold cross-validation) while keeping the number of epochs for the Multilayer Perceptron as 3. Keep all the other parameters at their default settings.What accuracy do these methods achieve on "bank-data.arff" and "mnist-small.arff"? Can you think of reasons to explain their relative accuracy on these data sets? What are the properties of the data sets? (Note: By default Weka takes the last column in the input arff file to be the target class, but this is not the case in "mnist-small.arff".)

Question 2

Apply both J48 and MultilayerPerceptron on the following data sets.

• vote.arff
• unbalanced.arff
• supermarket.arff
• mnist-large.arff

Our interest is the time taken to build the decision tree and neural network models in each case. Since the experiments could take time, keep the split percentage as 80, and the number of epochs for the MultilayerPerceptron as 1 for "mnist-large.arff". Plot a common graph of time (for building the model, not testing) versus the size of the data set, with one curve for each classifier.