SMOTE - Handle imbalanced dataset | Synthetic Minority Oversampling Technique | Machine Learning

[@mahipatil369]

Can i use smote technique for handling imbalanced data where classes are 3. While doing sentiment analysis using basebert model ?

[@hafizhaaghniahasya358]

Is it okay to use smote after feature encoding using OHE?

[@akshaypatil8155]

4:05 if it's a high dimensional data what technique should be used both for undersampling and oversampling?

[@DataMagicAI]

SMOTE is good choice for over sampling. For under sampling you can go with random under sampling.

[@Bumbazz]

how can we balance a multiclassed dataset?

[@akshaypatil8155]

for SMOTE technique is it necessary to have all the input columns and target variable in numerical format? or is it okay if some of the input columns are categorical?

[@DataMagicAI]

You can use Synthetic Minority Over- sampling Technique-Nominal Continuous (SMOTE-NE) instead of just SMOTE when you have numerical as well as categorical features.

[@MargaritaHantke]

Why can't you oversample or undersample just a bit more? Why do they have to be equivalent in 50-50% proportion?

[@DataMagicAI]

Its is to reduce the bias. If you provide more sample from class A then there is high chance like trined model will learn more about class A and get biased towards it. Thats the reason we want to make all class sample size equivalent to avoid such bias Trined models. There also some models which are capable to handle the unbalanced dataset. So in that case we dont need to do balancing.

[@shrikant136661]

If this imbalance data set is sparse then SMOTE will not be usefull as it works on base of KN theory,then what to do in that case?

[@DataMagicAI]

Try for the data augmentation techniques with GAN.

[@tornyu]

Checking my understanding: so SMOTE uses KNN and _linear_ interpolation to generate synthetic data, is that right? But linear is a big assumption, so by using a GAN you hope to synthesise samples that are closer to being in-distribution?

[@Prasannashinde365]

how to buy google colab pro in india can you help me?

[@DataMagicAI]

colab.research.google.com/signup visit this link select plan free/ Pro/ Pro plus and sign up.