Exam Professional Data Engineer topic 1 question 157 discussion

Seems to be A. Preprocessing/scaling should be done with input features, instead of predictions (output)

A Deep LEarning is not always the best solution D talks about fudgin the output which is wrong

https://www.quora.com/How-can-I-improve-Precision-Recall-AUC-under-Imbalanced-Classification

B,C are simply not true. D is modifing the scoring, making it not realiable anymore. A makes sense, is potentially increasing the model accuracy.

a is the correct answer

Answer A, why not B, Deep Neu Net. are better for sure but AUC is 0.87 is already good. Don't go for complex and time taking model. If AUC more than 0.95, it can be a reason of overfit. Now just check SVM params for hypertuning if you can bring it close to 0,9-0,95

a is the correct answer

Also a good read is: https://cloud.google.com/ai-platform/training/docs/hyperparameter-tuning-overview

as mentioned by Spider7 "performing tuning rather than using the model default parameters there's a way to increase the overall model performance --> A."

Correct: A

0.89 it's already not bad but by performing tuning rather then using the model default parameters there's a way to increase the overall model performance --> A.

Not C because real-world AUC value falls between 0.5 and 1.0 usually, this wouldn't help. A seems the most straigh forward.

For a large enough training set DNN will most likely beat a SVM. However the opposite may or may not be true. It also depends on the complexity of the problem. Which we don’t know from the question. For image, nlp, I say B can be a good answer However, if we decide to stick with SVM, D reduces overfitting and may increase AUC. I am torn between the two!

Ans - D when the model is overfitted means want to increase the AUC, we always perform hyperparameter tuning, Increase regularisations, decrease input feature parameters etc.

AUC is scale-invariant. It measures how well predictions are ranked, rather than their absolute values. So answer shall be A https://developers.google.com/machine-learning/crash-course/classification/roc-and-auc?hl=en

Definitely not D https://developers.google.com/machine-learning/crash-course/classification/check-your-understanding-roc-and-auc

A for me, read below link for more details. https://towardsdatascience.com/understanding-hyperparameters-and-its-optimisation-techniques-f0debba07568