Chevron Left
Retour à Effectively Dealing with Imbalance Classes

Avis et commentaires pour d'étudiants pour Effectively Dealing with Imbalance Classes par Coursera Project Network

À propos du cours

In this 2 hour guided project you will learn how to deal with imbalance classification problems in a profound manner, applying several resampling strategies and visualizing the effects of resampling on imbalance classification dataset. Note: This project works best for learners who are based in the North America region. We’re currently working on providing the same experience in other regions....
Filtrer par :

1 - 2 sur 2 Avis pour Effectively Dealing with Imbalance Classes

par Mohammed

29 nov. 2020

Learnt lot of new technique for handling imbalanced dataset. Good project. However, I strongly recommend downloading the dataset from internet and run locally. Coursera's platform for running the project is confusing.....Overall good project.

par Yaron K

28 août 2021

Pro: Covers various techniques for handling imbalanced data sets. With metrics and visualizations

Con:

- Audio is sometimes hard to hear. However the closed captions help. - The Rhyme platform is problematic. Eventually I downloaded the dataset and ran the notebook locally. - The complete notebook isn't available. I think it is better when you have a complete notebook and can concentrate on listening to the lecturer and annotating the notebook.

Other: - There is no theory, just a demonstration of various techniques for under and over sampling. On the whole I think this is positive. There are a lot of techniques. So first find the one that is relevant to your data-set - and then study it further on the Internet. - It seems that there have been changes in the imblearn library so sm = SMOTE(), x_sm, y_sm = sm.fit_sample(X,Y) doesn't work. You need to change it to fit_resample().

Conclusion: Would definitely take another project given by the lecturer. Mainly because of the emphasis on visualizations and metrics. With so many examples on the internet - you'll always eventually have an ML that runs. The real challenge is ensuring the results are fit for use.