Exam AWS Certified Machine Learning - Specialty topic 1 question 156 discussion

AC would be my answer. As half the stores have only been open for 6 months, no seasonality would be captured. The aggregation of the daily also removes trends we see during the week which is also not great when we are looking for the daily predicated sales figure

Missing data and achieving daily predictions with weekly data will be issues.

I would go for AD A : Many stores have been in business for < 6 months --> unable to capture seasonality D : Zero sales are also sales records and will result in bias if omitted.

Since half of the stores are 6 months old seasonality would be a problem for them. instead of omitting weeks with no sales could lead to bias, requesting zero entries will help in predicting better

A as missing seasonality is an issue for the majority of the stores. D as we need to impute zeros as we would otherwise miss data. C won't do anything on performance.

The factors that will adversely impact the performance of the forecast model are: Sales data is aggregated by week. This will reduce the granularity and resolution of the data, and make it harder to capture the daily patterns and variations in sales volume. The data scientist should request daily sales data from the source database to enable building a daily model, which will be more accurate and useful for the prediction task. Sales data is missing zero entries for item sales. This will introduce bias and incompleteness in the data, and make it difficult to account for the items that have no demand or are out of stock. The data scientist should request that item sales data from the source database include zero entries to enable building the model, which will be more robust and realistic

C. Aggregated Weekly Data: Since the objective is to predict daily sales volume, weekly aggregated data might mask important daily trends and variations. Requesting daily sales data will provide a finer granularity of information that is crucial for building an accurate daily sales prediction model. D. Missing Zero Entries for Item Sales: The omission of weeks with no sales can lead to biased predictions, as the model might not correctly account for periods of no sales. Including zero entries for item sales would provide a more accurate representation of sales patterns, including the absence of sales, which is valuable information for the model. Based on this analysis, the factors that would most adversely impact the model's performance are the aggregated weekly data (Option C) and the omission of weeks with no sales (Option D).

A - six months is likely not enough to detect clear seasonality C - Can do weekly from daily but cant reliably do daily from weekly

Letters A and C are correct: we want to do a daily model (our base is on weeks) and we need to deal with new stores VS old stores. It is important to emphasize that the letter D also makes sense: we need to know the days when there were no sales, however the way it is written means saving lines (days of sales) with zero in the database, which is not practical.

the two factors that will adversely impact the forecast model's performance are seasonality detection for new stores and the aggregation of sales data on a weekly basis. The data scientist should request categorical data to relate new stores with historical data and request daily sales data from the source database to build a daily model, respectively, to mitigate these issues effectively.

AD. rest makes no sense.

A. Since more than half of the stores have been in business for less than 6 months, it will be challenging to detect seasonality patterns for these new stores. Therefore, one solution is to request categorical data to relate new stores with similar stores that have more historical data. This will help the model to identify common patterns and accurately forecast sales for new stores. C. Since the sales data is aggregated by week, it may not be possible to identify daily patterns or trends. Hence, one solution is to request daily sales data from the source database to enable building a daily model. This will help the model to identify daily patterns and improve its forecasting accuracy.

I go with CD. How could we ignore the days with 0 sales? The model should be trained so that it can predict 0 sales days as well.

B, C, D are possible. A couldn't be an answer because the model must predict daily sales volumes while A says 'Request categorical data'.