Weka: Predicted values

0
votes

In Weka, I am using the J48 classifier to make predictions. However, I notice that in the prediction output the instance order (of both actual and predicted instance values) does not correspond to the original order (ie. rows) of the dataset that I am working with. Any idea what might be causing this?

1
csvinstancewekapredictionj48
I just noticed that this is due to the selection of the cross-validation test option; using the training set will not affect instance order of actual and predicted values. – AppleCinnamon
Please give us an example. I understood that result rows are in different order than dataset. – bolec_kolec
Correct, and this results from the cross-calidation option. I have found the answer to my question, so this is solved. – AppleCinnamon

1 Answers

0
votes

I think it is caused by feature selection method using to construct decision tree. when a feature was selected at root of tree,it is possible that training data split to disjoint set of entities and maybe all of entity in one subset have same class. so this class appear early in confusion matrix.