Google Professional Machine Learning Engineer Question 12
Single answerGoogle Cloud PlatformYou are working for a retail company to forecast monthly sales for thousands of products across multiple stores. The dataset is a time-series with features such as product category, store location, historical sales, and promotional data. You need to build a machine learning model that accounts for seasonality, handles missing data, and provides interpretable results. Which of the following approaches should you prioritize?
- A
Use a boosted trees model like XGBoost with engineered features for seasonality and promotions.
- B
Apply matrix factorization to decompose the data into latent factors for products and stores.
- C
Train an autoencoder to identify and reconstruct patterns in the time-series data.
- D
Fit a regression model with manually engineered features for seasonality and trends.
Show answer and explanation
Correct answer: A
Explanation
The boosted trees approach is the most suitable for this scenario as it can handle the complexity of the dataset, capture interactions between features, and account for missing data. Additionally, by engineering features to represent seasonality and promotions, it can model the time-series effectively while still providing interpretable results. Other methods like matrix factorization or autoencoders are not well-suited for this type of forecasting, and regression models may require more manual work and still underperform compared to boosted trees.
- A. Correct.
Boosted trees models like XGBoost are highly effective for tabular data with complex interactions. By engineering features for seasonality and promotions, you can capture temporal patterns and make accurate predictions. Additionally, boosted trees inherently handle missing data well, making it a strong choice for this scenario.
- B. Incorrect.
Matrix factorization is typically used for collaborative filtering tasks, like recommendation systems, where the goal is to learn latent factors from sparse data. It is not well-suited for time-series forecasting or for handling seasonality and promotions explicitly.
- C. Incorrect.
Autoencoders are primarily used for dimensionality reduction or anomaly detection. While they can model patterns, they are not inherently interpretable and are not the most efficient method for time-series forecasting in this case.
- D. Incorrect.
Regression models can be used for time-series forecasting, but they often require significant manual feature engineering. Additionally, they may not perform as well as boosted trees for complex datasets with non-linear relationships.