Scribd
Upload
40 views6 pages

W7 Weka Experimenter

The document discusses using the Weka tool to perform classification experiments on diabetes and weather datasets. It describes setting up experiments using different classifiers like Naive Bayes, J48, and OneR. For the diabetes data with cross-validation, Naive Bayes had the best performance. When changing to training/percentage split, the results were less accurate. For the weather data using cross-validation, Naive Bayes again performed best with results similar to its cross-validation output, while ZeroR and OneR results differed more from their cross-validation outputs. Using training/percentage split brought ZeroR closer to its classify output.

Uploaded by

Azfar Jiji
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
40 views6 pages

W7 Weka Experimenter

The document discusses using the Weka tool to perform classification experiments on diabetes and weather datasets. It describes setting up experiments using different classifiers like Naive Bayes, J48, and OneR. For the diabetes data with cross-validation, Naive Bayes had the best performance. When changing to training/percentage split, the results were less accurate. For the weather data using cross-validation, Naive Bayes again performed best with results similar to its cross-validation output, while ZeroR and OneR results differed more from their cross-validation outputs. Using training/percentage split brought ZeroR closer to its classify output.

Uploaded by

Azfar Jiji
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 6

W7 BI tuto Lab Work: Weka Classification using Experimenter

1. Launch the WEKA tool, and activate the Experimenter environment.


a. On the Setup Tab, Click New to start on new experiment.
b. On the Dataset click Add new to add the “Diabetes. arff” dataset (from Weka/Data)
c. Observe the dataset and perform the tasks to answer the questions:

i. Identify the total attributes used to represent the instances?

-9 attribute

ii. Which attribute is the class label and what are the possible identified class?

- class(nominal)

iii. What are the data type of each of the attributes in the dataset?

-preg(numeric),plas(numeric),skin(numeric),insu(numeric),mass(numeric),pedi(numeric),age(numeric),

class(nominal)

iv. Go to the the Experiment Type section and set it as Cross Validation. Set the Iteration

Control Section to repeat for 10 times, and set prioritize the data set first.

v. Set the Algorithm section and click the Add New button to add the following algorithms, click

Choose button:

i. Select Bayes and the NaiveBayes as the first classifier.

ii. Select Rules and the OneR Classifier

iii. Select Tree and J48 Classifier

d. Go to Run tab click Start to perform the test for each of the classifier

e. Go to Analyse tab and at Source Section click Experiment button and next at the Action Section

Click Perform Test button


f. Observe the results and propose the best model. Justify your answer.

-the best model is bayes.Na because it significantly more accurate

g. Try to change the experiment type to Train/Percentage Split (data Randomize). Run and

Perform the test. Compare the results to the previous experiment. Explain the result.

-the data is less accurate


2. Repeat the test using Experimenter by using the WeatherNominal.arff dataset in Weka. This time

use (3) three classifiers as follows: ZeroR, OneR and NaiveBayes and compare the results

a. Use the “Cross-validation” (10 folds) test mode

b. Observe the results shown in the “Classifier output” window for each of the algorithm

applied.
c. How many instances are incorrectly classified?

-(refer the three images above for each individual algorithm)

d. Compare these results with those observed for the each of the classifier in the crossvalidation test
mode.

-ZeroR in classify output is 64% which is lower than its crossvalidation test(70)

- OneR in classify output is 42% which is higher than its crossvalidation test(38)

- NaiveBayes in classify output is 57.1% which is similar to its crossvalidation test(57.5)

e. Which classifier, ZeroR, J48 or NaiveBayes, shows a better prediction performance for the

current dataset and the cross-validation test mode

-NaiveBayes has better prediction since its classify output

f. Try use training /Percentage split for the experiment type. Perform the test and explain if

there are any difference on the models.


-yes there is a difference where ZeroR is signifacantly closer to its classify output

You might also like