QuestionQ63

VerifiedML Model Development

A company uses Amazon SageMaker AI to build a classification model that rates the company’s monthly sales performance for each month over the past 20 years on a scale of 1 to 5. The dataset contains fields for month, sales region, regional aggregate sales, and the number of stores in each sales region. The company observes that aggregate sales values are unexpectedly high during two months of every year. The company applies one-hot encoding to every non-numerical feature in the training and validation datasets. The company trains the classification model with the training dataset. When the company evaluates the model using the validation dataset, the results are less accurate than anticipated.

The company must increase the model’s accuracy on the validation dataset.

Which solution will meet this requirement?

Explanation

Aggregate sales should be normalized independently for each sales region because regions can have different baseline sales ranges. Per-region normalization preserves recurring high-sales periods while expressing each value relative to that region’s distribution, preventing larger-volume regions from disproportionately influencing the classifier. AWS documentation describes normalization as feature scaling and notes that it prevents a model from being dominated by the weight of a single feature.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!