Multi class classification



I have a Multi class classification problem with 3 classes.Class 1 is being differentiated from other classes but I have a problem separating Class 2 from Class 0.(Most of the class 2 is being predicted as Class 0).

Since there is a imbalance between these two classes (3:1) .I tried giving weights to Class 2 but then the model just classifies some of the Class 2 correctly but it also predicts Class 0 as Class 2 now, giving me the same accuracy when where was no weights given.

So any ideas on how to separate these two classes? I have tried NN, oneVSrest, also tried a bit of stacking with no significant gain.



If you are facing problem in dealing with multiclass target variable, you can perform one-on-one classification, that is, you can keep class 1 as class 1 and combine class 2 and class 3 as class 2.
In the next case you can keep class 2 as one and and the rest as another.