Exam AWS Certified Machine Learning - Specialty topic 1 question 109 discussion

I think it should be C as the final outcome among 200 categories is already know. No need to build a classification model. It's pure forecasting problem.

C is my answer. No need to do classification. Because you know whether the insurance has a claim or not in the dataset. The claim contents do not provide additional information.

forcasting

It's pure forescasting problem.

I would say no machine learning model needed at all. Just using count group by categories SQL is enough

FinalOutcome 1 2 . 200 RecordID, FinalOutcome, Date, ClaimContents 1 2 . 100000 Note: claim content has partial information, only for few of 200 categories predict how many claims to expect in each category from month to month, a few months in advance We dont need the claim contents, we have all we need from first 3 columns to train a forecast model c

Forecasting using claim IDs and timestamps to identify how many claims in each category to expect from month to month. The problem requires the prediction of the number of claims in each category for each month, which is a time series forecasting problem. The timestamps and record IDs can be used to model the underlying patterns in the data, and the model can be trained to predict the number of claims in each category for future months based on these patterns. While the claim contents might provide additional information, the fact that partial information is only available for a few categories suggests that this information might not be enough to build a robust model, and that it might not be possible to apply supervised learning to all 200 categories. Instead, the model should be trained on the time series data (claim IDs and timestamps) for all categories, and the claim contents can be used to improve the accuracy of the model only for the categories for which such information is available.

how can a forecasting/classification model can be based on the claim ID? (that should be unique)

it's a forecasting problem, not a classification one

predict how many claims to expect in each category from month to month, a few months in advance C is the only one mentioning forecasting

D is correct. Multi-label classification to impute the missing claim contents, then forecasting what we want. C is missing the imputation part.

It is true that the final outcome is known. But C does not use the partial information from the 200 categories. Reinforcement learning currently is state of the art in stock prediction and other time series. Why waste valuable information? For me it's B.

This is a supervised learning approach: Supervised learning problems can be further grouped into regression and classification problems. Classification: A classification problem is when the output variable is a category, such as “red” and “blue” or “disease” and “no disease.” Regression: A regression problem is when the output variable is a real value, such as “dollars” or “weight.”