Assignment 1: Introduction to the WEKA Data Mining Software

Decision Trees

1.1 Study the animals in the Excel document (zoo.xls). Without using a data mining tool, draw a decision tree of three to five levels deep that classifies animals into a mammal, bird, reptile, fish, amphibian, insect or invertebrate.

1.2 Read about the ARFF-format here. Construct the header for the animal file.

1.3 Download datasets.zip and unzip it. Open zoo.arff by going to Weka and then choosing the explorer.

1.4 Find out in WEKA how many animals this dataset contains.

1.5 Go to the classifier tab and select the decision tree classifier j48. Click on the line behind the choose button. This shows you the parameters you can set and a button called 'More'. Which algorithm is implemented by j48?

1.6 Which percentage of instances is correctly classified by j48? Which families are mistaken for each other?

1.7 Again go to the parameter settings by clicking on the box after the 'Choose' button. Now change binarySplit to true and build a new decision tree. What is the difference?

1.8 Experiment with some of the other classifiers and until you get a better classification performance. Write down the classifier and its performance.

1.9 Compile the following source code in Java:

import java.io.*;
import weka.classifiers.trees.J48;
import weka.core.*;

public class MyDecisionTree {
MyDecisionTree(){
try{
FileReader reader = new FileReader("zoo.arff");
Instances instances = new Instances(reader);

// Make the last attribute be the class
instances.setClassIndex(instances.numAttributes() - 1);
J48 tree = new J48();
tree.buildClassifier(instances);
System.out.println("The third animal is classified as: " + tree.classifyInstance(instances.instance(2)));

reader.close();
} catch(Exception ex){
ex.printStackTrace();
}
}

public static void main(String args[]){
new MyDecisionTree(); }
}

To compile: javac -classpath weka.jar MyDecisionTree.java, note: first copy weka.jar and zoo.arff to the same directory as MyDecisionTree.java.

To execute: java -classpath .;weka.jar MyDecisionTree

(in some versions of java ; must be a : or it must be -classpath=.:weka.jar, please e-mail me if it doesn't work)

Use the Weka API documentation. How can you make a decision tree with a binary split?

Association rules

Next we will search for association rules.

2.1 The association algorithm requires nominal variables. In order to make all variables nominal we need to distcretize the data.

This pre-processing can be done with filtering, find the filter button on the pre-processing tab and select the right unsupervised method to convert the attributes to nominal attributes. After selecting a filter you can set its properties by clicking on it. Press the apply button and watch how the attributes change.

2.2 Now run the association rule algorithm. Which rules are always true? Write them down.

2.3 Write down a couple of interesting rules and a couple of trivial rules.

Pima indians, mushrooms and politicians

3.1 The dataset.zip file contains different data sets ranging from predicting diabetes in an indian population, distinguishing eatable mushrooms from poisonous till separating republicans from democrats. Most datasets contain a short description in the 'header'. Choose at least one data set, and answer the following questions:

• What needs to be predicted (i.e. which class)?
• Build a classifier and give the quality of the prediction.
• Give one or more interesting association rules and explain why these are interesting.

Other datasets

On the internet you can find many more data sets. Not all these data sets are in the ARFF format. Choose one of the data sets from http://kdd.ics.uci.edu/. Convert this dataset to the ARFF format and try different data mining techniques.