MLS-C01 Practice Exam 63 — 20 Questions

Question 1

A machine learning engineer must improve a SageMaker-built XGBoost model that currently overfits a high-dimensional sparse text feature set with 200,000 features and 20,000 documents. The engineer wants to reduce dimensionality and improve generalization. (Choose two.)

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use colsample_bytree to subsample a fraction of features per tree. Sparse, very high-dimensional feature sets let tree models carve out tiny leaves keyed to rare tokens. Raising min_child_weight blocks splits that create underpopulated leaves, and colsample_bytree restricts each tree to a random feature subset, both reducing variance. These two SageMaker XGBoost hyperparameters directly target the overfitting and dimensionality concerns described.

Answer

Increase the number of boosting rounds without any other change

Answer

Set the eval_metric to mape and select the best iteration by that metric

Answer

Switch the objective to multi:softmax

Question 2

A data scientist is using Amazon SageMaker to train a model with the built-in XGBoost algorithm. The training job is taking longer than expected, and the scientist wants to reduce the training time without significantly sacrificing model accuracy. Which two actions should the scientist take? (Choose two.)

Accepted Answer

Reduce the number of features by performing feature selection before training.. To reduce training time while preserving accuracy, the most effective actions are scaling compute resources (larger instance or distributed training) and reducing the number of features through feature selection. Both directly decrease the computational workload or increase parallel capacity without inherently degrading model quality. Increasing boosting rounds or using the exact tree method would lengthen training, and reducing max_depth may hurt accuracy, so they are not the best choices.

Answer

Set the 'tree_method' hyperparameter to 'exact' instead of 'approx' or 'hist'.

Answer

Decrease the value of 'max_depth' to a smaller number.

Answer

Increase the number of boosting rounds (num_round) significantly.

Question 3

A machine learning team is using Amazon SageMaker to train a deep learning model for image segmentation. They notice that the model's performance on the validation set is much worse than on the training set, even after adding dropout and data augmentation. They want to further reduce overfitting without collecting more data. Which technique is MOST likely to help?

Accepted Answer

Apply weight decay (L2 regularization) to the convolutional and fully connected layers.. When dropout and data augmentation have not sufficiently reduced overfitting, adding weight decay (L2 regularization) is a direct way to penalize complex weights and improve generalization. It constrains the model's effective capacity without new data. Increasing epochs or merely lowering the learning rate does not address the underlying complexity, and batch normalization is not primarily an overfitting remedy. Weight decay is a well-established, effective next step.

Answer

Increase the number of epochs.

Answer

Add batch normalization layers.

Answer

Reduce the learning rate and train for more epochs.

Question 4

A team trains a SageMaker built-in Linear Learner model for a regression task predicting delivery time in minutes. On the holdout set the model performs well on average but systematically under-predicts very long deliveries. The team wants to change the training so that large errors are penalized more heavily, accepting that small errors matter less. Which hyperparameter change should they make?

Accepted Answer

Set the loss hyperparameter to squared_loss instead of absolute_loss.. The built-in Linear Learner supports absolute_loss and squared_loss for regression. Squared loss grows with the square of the residual, so it assigns disproportionate cost to the large errors seen on long deliveries, pushing the model to reduce them. Changing predictor type, batch size, or label normalization does not alter the asymmetric penalty the team needs.

Answer

Increase the mini_batch_size used during training.

Answer

Set predictor_type to binary_classifier.

Answer

Enable the normalize_label option on the target column.

Question 5

A data scientist is using Amazon SageMaker to train a model with the built-in Object2Vec algorithm. The goal is to learn low-dimensional embeddings for a set of products based on customer purchase sequences. The dataset contains millions of purchase sequences. The data scientist wants to ensure the model generalizes well to unseen products. Which hyperparameter is MOST important to tune to prevent overfitting?

Accepted Answer

The dropout rate. In neural network-based algorithms like Object2Vec, dropout is a key regularization technique to prevent overfitting. It randomly deactivates neurons during training, forcing the network to learn more robust features. For generalization to unseen products, tuning dropout is more effective than adjusting epochs, batch size, or learning rate.

Answer

The batch size

Answer

The number of epochs

Answer

The learning rate

Question 6

A data scientist is training a gradient boosting model using SageMaker's built-in XGBoost algorithm. The model is overfitting on the training data. Which hyperparameter adjustment is most likely to reduce overfitting?

Accepted Answer

Increase lambda (L2 regularization). Increasing lambda (L2 regularization) penalizes large leaf weights in XGBoost, shrinking the model's complexity and reducing its ability to fit noise in the training data, which directly combats overfitting. L2 regularization is one of the primary levers in the XGBoost objective function alongside alpha (L1) and gamma (minimum loss reduction). Raising lambda is a standard, low-risk adjustment when a boosted model overfits.

Answer

Increase learning rate (eta)

Answer

Increase max_depth

Answer

Increase num_round

Question 7

A team is training a binary classifier and obtains a confusion matrix with 100 true positives, 10 false positives, 20 false negatives, and 200 true negatives. What is the precision of the model?

Accepted Answer

0.91. Precision is calculated as TP / (TP + FP). With 100 true positives and 10 false positives, precision = 100 / (100 + 10) = 100 / 110 ≈ 0.909, which rounds to 0.91. This metric measures how many of the positive predictions were actually correct.

Answer

0.87

Answer

0.94

Answer

0.83

Question 8

A data scientist needs to perform hyperparameter optimization for a model. Which AWS service provides built-in hyperparameter tuning jobs?

Accepted Answer

Amazon SageMaker. Amazon SageMaker provides built-in hyperparameter tuning jobs as a managed service, allowing data scientists to automatically search for optimal hyperparameter values using strategies like Bayesian optimization, random search, or Hyperband. This is a core feature of SageMaker's automatic model tuning capability, which integrates directly with SageMaker training jobs and supports early stopping to reduce compute costs.

Answer

Amazon EMR

Answer

AWS Step Functions

Answer

AWS Batch

Question 9

A data scientist is running exploratory data analysis in a SageMaker Studio notebook on a 12 GB CSV file stored in Amazon S3. The notebook kernel keeps running out of memory when a pandas DataFrame is loaded with pd.read_csv(). The scientist needs to compute column-level summary statistics and correlation values without downloading the entire file into memory. Which approach should the scientist use?

Accepted Answer

Load the file with Dask or PySpark using chunked or distributed reads, then compute describe() and corr() on the resulting collection.. Chunked or distributed readers such as Dask and PySpark keep the working set bounded by partition size rather than total file size, which is exactly what the notebook needs. They also ship describe() and corr() equivalents, so the scientist gets the same column statistics and correlation values that pandas would produce, but computed partition-wise. Larger instances or columnar formats still require materializing the full table in the kernel.

Answer

Increase the Studio notebook instance to a larger memory-optimized instance and rerun pd.read_csv() on the full file.

Answer

Convert the CSV to Apache Parquet with AWS Glue, then call pandas.read_parquet() to load the full dataset before computing statistics.

Answer

Use AWS Glue DataBrew to create a profile job against the S3 prefix and read the generated statistics from the job output.

Question 10

A data scientist is working with a dataset that contains a 'Price' column. After plotting a histogram, they observe that the distribution is right-skewed with many extreme high values. They plan to use a linear model that assumes normally distributed errors. Which of the following transformations should they apply to the 'Price' column to make it more normally distributed?

Accepted Answer

Apply log transformation (log(Price)).. Log transformation is commonly applied to right-skewed data to reduce skewness and make the distribution more normal, which is suitable for linear models assuming normally distributed errors. Option B (square transformation) exacerbates skewness, making it worse. Option C (min-max scaling) only rescales the data to a fixed range and does not change the shape of the distribution. Option D (binning) discards information and does not transform the distribution to be normal.

Answer

Apply square transformation (Price^2).

Answer

Apply min-max scaling to the 'Price' column.

Answer

Bin the 'Price' values into equal-width intervals.

Question 11

A data scientist is exploring a marketing dataset in Amazon SageMaker Studio and wants to identify which of 60 numeric features are most strongly associated with a binary conversion label, while ignoring any pair of features that are nearly redundant with each other. The dataset fits comfortably in memory. Which combination of techniques is most appropriate for this exploratory step?

Accepted Answer

Compute point-biserial or mutual-information scores between each feature and the binary label, then drop one feature from each pair whose absolute Pearson correlation exceeds a threshold.. Because the label is binary and the predictors are continuous, point-biserial correlation or mutual information is the appropriate association measure, with mutual information adding sensitivity to nonlinear relationships. Pairwise correlation filtering then removes near-duplicate features so the importance ranking reflects distinct signals. Regression, clustering, and PCA each mismatch the scenario by assuming the wrong target type or by ignoring the label entirely.

Answer

Fit a linear regression on the label and inspect the p-values of each coefficient.

Answer

Apply principal component analysis and select the components with the largest explained variance as the most predictive features.

Answer

Run k-means clustering with k equal to the number of features and inspect cluster centroids to rank feature importance.

Question 12

A data scientist is performing EDA on a dataset with a highly skewed numerical feature. They want to apply a transformation to make the distribution more symmetric before modeling. Which transformation is most appropriate?

Accepted Answer

Log transformation. A log transformation is the most appropriate for reducing high right skew because it compresses large values more than small ones, making the distribution more symmetric. While Box-Cox is more flexible, it requires tuning and positive data; log is simpler and often effective. Square root is too mild, and standardization does not alter distribution shape.

Answer

Square root transformation

Answer

Box-Cox transformation

Answer

Standardization

Question 13

A data scientist is working with a dataset containing 10,000 observations and 100 features. The scientist wants to detect outliers in the dataset. Which method is most appropriate for outlier detection in a high-dimensional space?

Accepted Answer

Use Isolation Forest. Isolation Forest is an ensemble method specifically designed for anomaly detection in high-dimensional datasets. It isolates observations by randomly partitioning the data; outliers require fewer splits to isolate, so they have shorter path lengths in the trees. This makes it robust to the curse of dimensionality and does not assume a particular data distribution, unlike distance- or density-based methods.

Answer

Use Z-score to identify points beyond 3 standard deviations

Answer

Use Mahalanobis distance

Answer

Use interquartile range (IQR) for each feature

Question 14

A company is building a binary classifier to detect fraudulent transactions. The dataset is highly imbalanced with only 0.1% positive cases. The data scientist uses logistic regression and obtains 99.9% accuracy on the test set. Which metric should the data scientist use to evaluate the model's performance?

Accepted Answer

Precision-recall curve. With only 0.1% positive cases, accuracy is misleading because a model that always predicts 'not fraudulent' achieves 99.9% accuracy. The precision-recall curve focuses on the positive class and is robust to extreme class imbalance, showing the trade-off between precision and recall across thresholds. This makes it the best choice for evaluating a binary classifier on highly imbalanced fraud detection data.

Answer

ROC AUC

Answer

Precision

Answer

F1 score

Question 15

A data scientist is exploring a dataset of 80,000 customer records in Amazon SageMaker Data Wrangler. The 'region' column has 6 categories, the 'plan_type' column has 3 categories, and the target 'churned' is binary. The scientist wants to visually determine whether churn rate varies across combinations of region and plan_type before engineering interaction features. Which Data Wrangler analysis best supports this?

Accepted Answer

A Categorical Target Distribution analysis with churned as the target and both region and plan_type as grouping columns.. The scientist needs churn rate broken out by the cross of two categorical features. Data Wrangler's Categorical Target Distribution analysis takes a target and grouping columns and reports the target's distribution within each group, so using churned with region and plan_type surfaces the two-way interaction pattern. Histograms, scatter plots, and column summaries do not cross-tabulate a binary target by two categoricals.

Answer

A Scatter Plot analysis with region on the x-axis and plan_type on the y-axis.

Answer

A Histogram analysis on the churned column, grouped by plan_type.

Answer

A Multicolumn Summary analysis across region, plan_type, and churned.

Question 16

A data engineer must run exploratory analysis on a 2 TB dataset of JSON event logs in Amazon S3. The logs are newline-delimited JSON with inconsistent optional fields, and the engineer needs ad hoc SQL queries plus automatic schema discovery without managing any servers. Which solution meets these requirements with the least operational effort?

Accepted Answer

Create an AWS Glue crawler to populate the Data Catalog, then query the logs with Amazon Athena.. A Glue crawler plus Athena is the minimal-effort combination because both are serverless. The crawler handles schema inference and partition registration for semi-structured JSON, and Athena queries the catalog tables in place without copying data. EMR and Redshift add cluster management or data loading, while the Parquet conversion job adds an ETL pipeline that is unnecessary for exploratory querying.

Answer

Use AWS Glue Studio to build an ETL job that writes the logs to Parquet, then query with Amazon Athena.

Answer

Launch an Amazon EMR cluster with Hive, define an external table over the S3 prefix, and query with HiveQL.

Answer

Load the JSON logs into Amazon Redshift with COPY and query them with Redshift SQL.

Question 17

A data scientist is analyzing a dataset with missing values. The missing data mechanism is missing at random (MAR). Which imputation method is most appropriate to preserve relationships between variables?

Accepted Answer

Use multiple imputation by chained equations (MICE).. MICE (Multiple Imputation by Chained Equations) is the most appropriate method for MAR data because it models each variable with missing values conditional on the other variables, preserving multivariate relationships and injecting proper uncertainty across multiple imputations. It produces unbiased estimates under MAR and is the gold standard for complex missingness patterns.

Answer

Remove all rows with any missing values.

Answer

Use k-nearest neighbors imputation.

Answer

Replace missing values with the mean of the column.

Question 18

A data scientist is profiling a 200-column tabular dataset in Amazon SageMaker Data Wrangler before training. Several numeric columns contain extreme values, one column has 60% missing entries, and the target is continuous. The scientist wants to use Data Wrangler's built-in analyses to identify data-quality problems that could bias the model, without writing custom code. (Choose two.)

Accepted Answer

Run a Multicolumn Summary analysis to review per-column statistics such as missing counts, min, max, and quantiles across the 200 columns.. Two built-in analyses cover the stated problems. Multicolumn Summary reports per-column statistics, including missing counts and distributional extremes, so it finds both the 60% missing column and the outlier-heavy columns. Target Leakage compares each feature against the continuous target and flags suspiciously predictive columns that would bias evaluation. Quick Model, scatter plots, and target-only histograms do not report missingness or per-column outlier magnitude.

Answer

Run a Scatter Plot analysis on the continuous target to identify which numeric columns have extreme values.

Answer

Run a Quick Model analysis and use the reported feature impact to detect which columns contain missing values.

Answer

Run a Histogram analysis on the target column only, since target distribution determines whether feature outliers matter.

Question 19

A data scientist is performing exploratory data analysis on a dataset stored in Amazon S3 and wants to use Amazon SageMaker Data Wrangler to identify and handle missing values. The scientist needs to generate a report that shows the percentage of missing values per column and also wants to impute missing values using the mean for numerical features. Which two actions should the scientist take in Data Wrangler to achieve these goals? (Choose two.)

Accepted Answer

Apply the 'Handle Missing Values' transform and select the 'Impute' option with mean strategy for numerical columns.. To identify and handle missing values in SageMaker Data Wrangler, the scientist should first generate the Data Quality and Insights Report to see missing value percentages per column, and then apply the Handle Missing Values transform with the mean imputation strategy for numerical features. These two steps provide both the diagnostic view and the corrective action.

Answer

Use the 'Data Quality and Insights Report' to automatically impute missing values with the mean.

Answer

Use the 'Data Quality and Insights Report' to generate a histogram of missing values.

Answer

Apply the 'Handle Missing Values' transform and select the 'Drop' option for all columns.

Question 20

A machine learning specialist is performing exploratory data analysis on a dataset with a continuous target variable. The specialist suspects that the relationship between a key numerical feature and the target is not linear but monotonic. The dataset also contains several categorical features. Which visualization technique is most appropriate to assess the monotonic relationship between the numerical feature and the target while also allowing for the inspection of potential interactions with a categorical feature?

Accepted Answer

A line plot of the mean target value across bins of the numerical feature, with separate lines for each level of the categorical feature.. Binning the numerical feature and plotting the mean target per bin, with separate lines for each category, directly visualizes the functional form of the relationship. Monotonicity is evident if the mean target consistently increases or decreases across bins. Separate lines reveal whether the relationship differs by category, indicating interactions. This method is robust to noise and clearly communicates trends.

Answer

A heatmap of the correlation matrix between all numerical features and the target.

Answer

A scatter plot of the numerical feature versus the target, with points colored by the categorical feature.

Answer

A box plot of the numerical feature grouped by the categorical feature, with the target represented by the size of the boxes.