Data Mining With Weka - Experiment (Irish Dataset) & J48

Weka - Data Mining Tool

  Experiment  (Irish Dataset):

In this experiment I am using Irish dataset and different algorithm to show classification using 10 fold cross-validation methods, there will be 10 repetitions on the processes to determine the results.(Note: Use of all other datasets and algorithms is similar process). For beginners, start Weka and click on Experimenter Option

1  

     Using J48

     Here I am using J48 algorithm to Irish datasets, the process is as follows:

There are three panels starting with Setup Panel:

·         Click New to start new experiment

·         Click add new under datasets in order to add new dataset i.e. Irish.arff

·           Click add new under Algorithm in order to add new dataset i.e. J48

·         Experiment type is cross validation by default 

·         And Its classification by default

We can also choose other experiment types such as percentage split etc., regression types and we can set the number of repetitions under the Iteration control.

                               

Run Panel

·         Click start to run the setup algorithm and datasets

Furthermore, you can click stop to abort the process.  This is the log window that contains starting time as 23:14:09 i.e. after you click start and finished time 23:14:09 i.e. the process is finished. And there are no errors and status as not running. Overall, The J48 takes no time (negligible time span) and shows the experiment (classification) results without any errors.

 

  

Analyse Panel:

·         Click Experiment to get last run experiment

·          Select Options eg. standard deviations if you want on your result

·         Click Perform Test to get result summary or test output

Moreover, you can found available result list under the result lists and you also can save the output.

And significance is 0.05 or 5 percentage by default and v represents significantly better result while * represents significantly worse results based on value of significance.

Evaluating J48 on Iris dataset:  We got average of 95.33 percent correct using J48 on iris dataset. It’s a  10 fold cross validation so, if we want to see individual result we can save result on CVS file from setup panel.