classification results are good or not?

Question

[rajeshkumar_vinod]

Hello

i am new to weka classification so i wanted to ask if these results are good or not and if there is class imbalance also. i am getting following results from 10 fold cross validation using weka -

=== Stratified cross-validation ===

=== Summary ===

Correctly Classified Instances 338 (83.4568 %)

Incorrectly Classified Instances 67 (16.5432 %)

Kappa statistic 0.359

Mean absolute error 0.1654

Root mean squared error 0.4053

Relative absolute error 70.02 %

Root relative squared error 118.3005 %

Total Number of Instances 405

=== Detailed Accuracy By Class ===

             TP Rate  FP Rate  Precision  Recall   F-Measure  MCC      ROC Area  PRC Area  Class
             0.886    0.491    0.920      0.886    0.902      0.362    0.727     0.924     classA
             0.509    0.114    0.412      0.509    0.455      0.362    0.766     0.350     classB

Weighted Avg. 0.835 0.440 0.851 0.835 0.842 0.362 0.732 0.846

=== Confusion Matrix ===

a b <-- classified as

310 40 | a = classA

27 28 | b = classB

thank you.

Answer 1

[Philipp Bayer]

Not sure whether this is a bioinformatics question, not sure what your classes are - you definitely got class imbalance as classA seems to be about 10 times larger. If you don't have a baseline as to how good your classification rates are it's always good to try with another class where you randomly assigned items, asking whether your lass assignment is better than random assignment.

Look at methods to work around class imbalance such as https://github.com/scikit-learn-contrib/imbalanced-learn

Also try to see whether a simple logistic regression can outperform your method, that should be your base line to improve upon, such as http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html