Exam AWS Certified Machine Learning - Specialty topic 1 question 271 discussion

Option A allows the ML specialist to add class weights to the MLP’s loss function, and then retrain. Class weights are a way of assigning different importance or penalties to different classes in a classification problem. Class weights can help balance the data distribution and reduce the bias towards the majority classes. Class weights can also help improve the recall metric, which is the ratio of true positives to the sum of true positives and false negatives. Recall measures how well the model can identify the relevant instances of a class, especially when the class is rare or unique. The ML specialist can use class weights to increase the importance or penalty of the target class of interest, and then retrain the MLP to improve its recall.

Option A - Add class weights to MLP's loss function - improve recall with the least amount of time and effort by making the model more sensitive to the underrepresented target class during training.

Apologies for the confusion but on second thoughts, A is the right answer as unique doesn't mean unknown and this is still a supervised learning problem. Adding weights to classes would even out the bias caused by unique class and improve recall as mentioned by other experts in this forum. Please ignore my previous comment. A is the correct option indeed.

Leaning towards C as the target class of interest is unique as compared to dataset (as given in this question). If the target class is unique / non-existing in data set then we are talking about unsupervised learning and k-means is a right fit so, option C seems to be more appropriate than option A. Adding weights may still not be able to solve the purpose as target class is not present in data set. It is almost an unlabeled data set if target class is unknown / unique as compared to existing classes in data set. Unlabeled data sets are better solved using unsupervised learning.

Agreed A

A as this is Faster solution.