Exam DP-100 topic 6 question 3 discussion

does this question and answer make sense? i dont have any idea at all. could any one kindly give explain

The best I could gather was that: they would like to do crowd segmentation which would help them target certain people for their ad campaigns, using clustering based on videos and audios of the people in the crowd. The question wants us to create an evaluation strategy for the models they created. In the problem description, they said they noticed 47 features were not performing rightly and they would engineer 10 independent features from them before retraining our model. This gives us the first answer "Add new features for retraining...".

A - This will be the first one. I think this is part of Cluster-then-classification model. Based on my exp, I will use cluster result as a new feature for later classification model, that's reason we say "Add new features for retraining supervised models". E makes sense to me as well, but have no idea for C

A,C,B could be

- Filter labeled cases for retraining using the shortest distance from centroids: Start by identifying the labeled cases that are closest to the centroids of their respective clusters. These would typically be the most representative samples of their classes and would form a solid base for initial model training. C- Evaluate the changes in correlation between model error rate and centroid distance: After retraining the model with the selected cases, evaluate how the model's error rate correlates with the distance of samples from the centroids. This will provide insights into how well the model is performing and whether samples farther from the centroids are more likely to be misclassified. E- Filter labeled cases for retraining using the longest distance from centroids: Based on the evaluation in step 2, it may be observed that samples farther from the centroids are not being accurately classified. To improve the model's performance on these cases, they should be included in the training set for retraining.

The three actions that should be performed in sequence to define an evaluation strategy for the crowd sentiment models are: C) Evaluate the changes in correlation between model error rate and centroid distance: This step involves evaluating the correlation between the model's error rate and the distance from the centroid. It helps in identifying if the model is overfitting or underfitting the data. B) Filter labeled cases for retraining using the shortest distance from centroids: This step involves filtering the labeled cases for retraining based on the shortest distance from the centroids. This helps in selecting the cases that are closer to the centroids and are more representative of the cluster. A) Add new features for retraining supervised models: This step involves adding new features for retraining supervised models. The new features can help improve the performance of the models and capture important information from the data. Therefore, the correct order of actions is C, B, A.

Did this appear in any of the previous exams?

I cannot really follow this case study overall ... After compare with all options, I think the answer is logically sound ... No other comments ...

the question is complicated but i say that's a comparisation between existing sound and new sound so the first thing 1) add new features , seconde 2)use correlation to now how much new and old feure are correlated 3)evaluate

I couldn't make head or tails of this question. Clueless....

no idea about this.