Data Mining - Practical: Using Naïve Bayes Classifier for Classification

This is a combined practical and exercise session, in which we have two tasks. First, we use Weka to run two different classifiers (Naïve Bayes and J48) and compare their performances. Second, we use Weka to help us to calculate and decide whether someone is likely to get sunburned or not as Naïve Bayes Classifier does. This will help us to familiarise ourselves with the theories on Naïve Bayes Classifier we have learned in today's lecture.

Step 1:Launching Weka and Loading Training Data

Launch Weka by clicking on: RunWeka.bat

Select ‘Explorer' from the list of Applications.

Select the ‘Preprocess' tab and click on ‘Open File'. Choose the file ‘sunburn.arff' which contains the training data set.

Step 2: Exploring Naïve Bayes Classifier

Select the ‘Classify' tab and make sure that "NaiveBayes" is chosen from the classifier list and "Use training set" is clicked as the test option. Note that attribute "burned" is chosen by default as the class attribute but the class attribute can be changed if needed.

Click ‘Start'will create a classification model/classifier from the training dataset. The classifier is listed in the Result list while the details about the classifier are displayed in the ‘Classifier output' window.

Step 3: Comparing Naïve Bayes Classifier with Decision Tree-based Classifier J48 Using 10-fold Cross-validation

First, you can easily run Naïve Bayes Classifier again with a different evaluation method. Select the "cross-validation" test option with 10 folds as default and click Start again. The classifier output is quickly replaced to show how well the learned model performs on the cross-validation.

Second, you can also easily run J48 with 10-fold cross-validation. Select the ‘Classify' tab and make sure that "J48" is chosen from the classifier list and Select the "cross-validation" test option with 10 folds as default and click Start again. The classifier output is quickly replaced to show how well J48 performs on the cross-validation.

Finally, compare the two sets of evaluation results.

Q1. Provide a summary of the comparison between the classifier outputs by Naïve Bayes Classifier and J48.

Step 4: Discovering Naïve Bayes Classifier

We now focus on discovering how the Naïve Bayes Classifier produces the output. We have just covered the Naïve Bayes Classifier in today's lecture. We now want to familiarise ourselves with the theories that we have learned in today's lecture by working through an example using Weka.

Select the ‘Preprocess' tab and go back to examine the training data set. Click on attribute burned in the attributes window on the left and you will see that the class attribute has two class labels "burned" and ‘none".

Q2. Calculate the probability estimates of these two class labels, i.e., P(burned = "burned") and P(burned = "none") based on their class distributions. Click on attribute burned in the attributes window will display the class distribution, i.e., 9/16 with class label "burned" while 7/16 with class label "none"

P(burned = "burned") =
P(burned = "none") =

Click "Edit" to view the sunburn relation table that shows the value distribution of each attribute given a class label. Click on attribute burned will sort the instances in the table in alphabetic order. For example, among the 9 instances with class attribute value "burned", 2/9 has hair = "red".

Q3. Calculate the conditional probability of each attribute on each class, e.g., P(hair = "blonde"|burned = "burned"). Note that you would have 6 + 6 + 6 + 4 = 22 conditional probabilities to calculate.

P(hair = "blonde"|burned = "burned") =
P(hair = "brown"|burned = "burned") =
P(hair = "red"|burned = "burned") =
P(hair = "blonde"|burned = "none") =
P(hair = "brown"|burned = "none") =
P(hair = "red"|burned = "none") =
...

Q4. Calculate the classification probability of each class, given hair = "blonde", height = "short", weight = "average", lotion = "none".

P(burned = "burned"|hair = "blonde", height = "short", weight = "average", lotion = "none)
=

P(burned = "none"|hair = "blonde", height = "short", weight = "average", lotion = "none)
=

Q5. Is the given person likely to get sunburned based on the prediction by Naïve Bayes Classifier?

Attachment:- Practical2.rar