Im currently learning Classification Neural Network and Im working with imbalance dataset. I have 2 model with 2 different result and 4 type of classes.
The 1st model there are no sampling or dimensionality reduction made with about 12000 data.
1st Model Model: NN Accuracy : 91%
The 2nd model, I perform ratio sampling, combine with RUS and SMOTE about total of 4000 (1000 data for each class). The majority sample about 8000 data has been reduced to 1000 data and minority data have been generated synthetically using SMOTE from 20 data to 1000 data.
2nd model Model: NN Accuracy: 89%
The question is, does the 2nd model valid with so much majority features has been reduced and minority features generated synthetically?
Updated results with CM measurement.
Thanks, K