How do you rectify the challenges posed by an imbalanced dataset?

Question

Anonymous · Accepted Answer

If the data set is the basis of a model which requires high precision then an
imbalanced data will result in unacceptable errors as the model will predict the
majority class every time. If the model is evaluated on just the accuracy score
it will look artificially good. Instead the Precision metric should be used.
Approaches to reduce the problem are: 1) Use the SMOTE approach to generate
synthetic examples of the minority class 2) under sampling the majority class.
Of the two approaches the first is preferable as under sampling the majority
class may alter the distribution of the data and may remove subsets. The
downside of SMOTE is that when it comes to the minority case the model will
overfit and will need additional checks that it can detect the minority case
when used on new data.

If the model is being developed with Python the imbalanced-learn package is
designed to work within the sklearn ecosystem to perform SMOTE.

Anonymous · Answer

I'll use resampling (oversampling/undersampling), SMOTE, adjust class weights,
and evaluate with F1-score or AUC-ROC instead of accuracy to handle imbalanced
dataset

Anonymous · Answer

We can consider various resampling techniques like Oversampling, undersampling
or SMOTE, so that we can remove the biasness in the distribution

How do you rectify the challenges posed by an imbalanced dataset?

Practice this question with AI

Go Premium

Community Answers

Unlock Full Access

Unlock Full Access

Unlock Full Access

Practice More Questions