ExamPark
Course menuMLA-C02

Exams / MLA-C02 / Study guide / Choosing ML models and algorithms

Choosing ML models and algorithms: MLA-C02 study notes and practice questions

What to know about Choosing ML models and algorithms for the MLA-C02 exam, from the ML Model and Foundation Model (FM) Development domain (24% of the exam guide), then 1 free question to test it. Part of the MLA-C02 study guide.

Last verified against current AWS documentation ·How we write our questions

What to know for the exam

These questions describe a business problem and its data, and ask which algorithm or SageMaker AI option fits.

Match the problem to an algorithm

Problem SageMaker AI built-in options
Classification or regression on tables XGBoost, Linear Learner, LightGBM, CatBoost, AutoGluon-Tabular
Forecasting many related time series with intervals DeepAR
Anomaly detection Random Cut Forest; IP Insights for unusual IP use
Clustering k-means
Dimensionality reduction PCA
Recommendations with sparse data Factorization Machines
Topic modeling LDA, Neural Topic Model
Text classification and embeddings BlazingText
Translation and summarization (sequence to sequence) Seq2Seq, or foundation models
Images Image classification, object detection, semantic segmentation
Nearest-neighbor lookup k-NN
Similarity between objects Object2Vec

Other routes

  • SageMaker JumpStart offers pretrained open models and solution templates to deploy or fine-tune.
  • SageMaker Autopilot (now in SageMaker Canvas) runs AutoML on tabular data and shows the candidate models.
  • Bring your own framework script (PyTorch, TensorFlow, scikit-learn) with script mode.
  • AWS AI services when a pretrained API already does the job.

Supervised or not?

  • Labels exist: supervised (classification, regression).
  • No labels, find structure: unsupervised (clustering, anomaly detection, PCA).
  • Learning from rewards: reinforcement learning.

Interpretability vs performance

Linear models and shallow trees are easier to explain; gradient-boosted trees and deep networks often score higher. Regulated decisions may favor simpler models or need SHAP explanations.

Common traps

  • Linear Learner for forecasting many time series. Use DeepAR.
  • k-means for labeled data. It’s unsupervised.
  • Random Cut Forest for forecasting. It detects anomalies.

Try the 1 practice question

Practice Choosing ML models and algorithmsTake the free practice exam

Question 1

Medium ·Link to this question (question 1)

A solar installer operates 2,000 residential systems across three states. For each site, it needs a daily forecast of energy output for the next 30 days, and it has three years of daily production history per site with strong seasonal patterns.

Many sites share similar weather and seasonal behavior, so the team wants one model that learns from all of them together. The forecasts must come with uncertainty ranges, and the team wants a built-in algorithm in SageMaker AI.

Which algorithm should the ML engineer choose?

Choose one answer.

  1. A. Linear Learner, trained on one row per site with the average output as its target.
  2. B. DeepAR, trained on all site time series together, which produces probabilistic forecasts for each series.
  3. C. K-means, to cluster the sites and then forecast each cluster with its average output.
  4. D. Random Cut Forest, trained on all site histories, to flag days whose output is unusual and use the flagged days as the forecast.
Show the answer and explanation

Answer: B

  • A. Wrong: Linear Learner treats rows independently and would average away the daily pattern and seasonality the forecast needs. It does not produce a forecast for each day with uncertainty ranges.
  • B. Right: Correct. DeepAR learns from many related time series at once and outputs probabilistic forecasts with prediction intervals. It handles seasonality across sites, which matches the requirement. Learn more: Use the SageMaker AI DeepAR forecasting algorithm (opens in a new tab)
  • C. Wrong: K-means finds groups but does not produce forecasts over time. Averaging clusters would also hide the daily pattern for each site.
  • D. Wrong: Random Cut Forest is an anomaly detector and does not produce future values. Flagging unusual days does not give the 30-day forecast the installer needs.

Explanation

DeepAR is a built-in algorithm for forecasting many related time series, and it outputs probabilistic forecasts that include intervals. Linear Learner loses the time structure, k-means does not forecast, and Random Cut Forest detects anomalies rather than predicting values.

Study

Want more on Choosing ML models and algorithms? The 5 full-length practice sets have 9 more Choosing ML models and algorithms questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets

Study Choosing ML models and algorithms

More ML Model and Foundation Model (FM) Development topics