https://jovian.ai/casella0798/badmodel I created the model above to predict red wine quality. I have 6 classes, from 3 to 8. Dataset is unbalanced, with a lot of classes 5 and 6. My model performs really bad, with a low accuracy, and a loss that does not decrease under 1. How can I improve this model? I tried changing #layers and #neurons, also the batch_size and the learning rate. Any suggestion, also related to hyperparameters optimization? Dataset is available here: https://archive.ics.uci.edu/ml/datasets/wine+quality it is the red version
$\begingroup$
$\endgroup$
11
-
$\begingroup$ how did you manage imbalance? $\endgroup$– Nikos M.Feb 20, 2021 at 18:23
-
$\begingroup$ I did not anything. What can I do? Something like SMOTE or undersampling? Or there is some trick to do on the NN? $\endgroup$– CasellaJrFeb 20, 2021 at 18:31
-
$\begingroup$ re-sampling is a possible solution to imbalance, else weight adjustment (where applicable) $\endgroup$– Nikos M.Feb 20, 2021 at 18:37
-
$\begingroup$ Are you treating this as an ordinal regression? $\endgroup$– DaveFeb 20, 2021 at 18:42
-
$\begingroup$ @Dave no I am treating as a multilabel classification using cross entropy loss $\endgroup$– CasellaJrFeb 20, 2021 at 18:44
|
Show 6 more comments