AI0-001 AI Models and Data Engineering Practice Test 2 — 15 Questions

Question 1

Refer to the exhibit. A stream processor ingests events. One event arrives with missing "user_id". What will happen?

Accepted Answer

The event will be rejected because user_id is required.. In stream processing systems like Apache Kafka or AWS Kinesis, if a required field such as 'user_id' is missing and the schema (e.g., Avro, JSON Schema) defines it as required, the event will be rejected at ingestion. The stream processor typically validates the event against the schema; if validation fails, the event is not accepted into the stream, preventing downstream processing errors.

Answer

The event will be stored in a dead-letter queue automatically.

Answer

The event will be accepted but user_id will be set to null.

Answer

The event will be accepted with a default user_id of 0.

Question 2

Refer to the exhibit. What is the recall of the model?

Accepted Answer

0.8. Recall is calculated as True Positives divided by (True Positives + False Negatives). From the confusion matrix, True Positives = 80 and False Negatives = 20, so recall = 80 / (80 + 20) = 80 / 100 = 0.8. This measures the model's ability to correctly identify all actual positive cases.

Answer

0.72

Answer

0.7

Answer

0.73

Question 3

A data scientist is preparing a dataset for training a classification model. The dataset has a column with missing values in 5% of rows. Which action should the data engineer take to minimize bias?

Accepted Answer

Impute missing values with the median of the column. Imputing missing values with the median of the column is a robust method that minimizes bias because the median is resistant to outliers and preserves the central tendency of the data. Removing rows with missing values can introduce bias if the missingness is not completely random, and replacing with a constant like 999 distorts the distribution and can mislead the model.

Answer

Remove all rows with missing values

Answer

Replace missing values with a constant such as 999

Answer

Use a model that can handle missing values natively

Question 4

A team is training a language model using a large text corpus. They want to ensure the model does not learn biased associations between gender and professions. Which data engineering technique should they apply?

Accepted Answer

Balance the representation of professions across genders. Balancing the representation of professions across genders in the training data directly addresses the root cause of biased associations. By ensuring that each profession appears with roughly equal frequency for all gender references, the model learns statistical correlations that are fair rather than skewed by imbalanced data. This is a fundamental data engineering technique for bias mitigation, as it prevents the model from encoding spurious correlations between gender and occupation.

Answer

Remove all gender-related words from the text

Answer

Use a pre-trained model that is already debiased

Answer

Apply adversarial debiasing during training

Question 5

A streaming data pipeline ingests sensor data from IoT devices. The data arrives at irregular intervals and contains occasional spikes. Which data transformation is most appropriate for preparing this data for a time-series model?

Accepted Answer

Downsampling to a fixed frequency using mean aggregation. Downsampling to a fixed frequency using mean aggregation is the most appropriate transformation because time-series models require uniformly spaced timestamps to learn temporal dependencies. By aggregating irregularly arriving sensor data into fixed intervals (e.g., every 5 minutes) using the mean, you smooth out occasional spikes and create a consistent input structure that models like ARIMA or LSTM can process without bias from irregular gaps.

Answer

Removing all rows with values outside 3 standard deviations

Answer

Using a sliding window to compute moving averages

Answer

Padding missing timestamps with zeros

Question 6

A deep learning model for image classification is overfitting due to a small dataset. The team decides to apply data augmentation. Which augmentation technique is least likely to preserve the label?

Accepted Answer

Random cropping and rescaling. Random cropping and rescaling is least likely to preserve the label because it can cut out the primary object or distort its proportions, potentially removing the discriminative features needed for correct classification. For example, cropping a dog image to show only the background or a leg could change the semantic meaning, making the label 'dog' incorrect. In contrast, other techniques like noise, rotation, or flipping typically retain the core subject and its label.

Answer

Adding random noise

Answer

Random rotation

Answer

Horizontal flip

Question 7

A data science team is building a model to predict customer churn. The dataset includes categorical variables like 'region' and 'subscription_type'. Which three preprocessing steps should be applied to these categorical features? (Select THREE).

Accepted Answer

Label encoding. Label encoding (B) is correct because it converts each categorical value in 'region' or 'subscription_type' into a unique integer, which is useful for tree-based models and for creating a compact numeric representation. Ordinal encoding (D) is correct because it maps categories to ordered integers when a meaningful rank exists (e.g., subscription tiers), preserving that order for the model. One-hot encoding (E) is correct because it creates a binary column per category, avoiding false ordinal relationships for nominal features like 'region' and is the standard approach for linear models and neural networks. Normalization (A) and standard scaling (C) are not appropriate here because they rescale continuous numeric values to a range or z-scores and do not transform categorical variables into a usable numeric form.

Answer

Normalization

Answer

Standard scaling

Question 8

A data engineer is designing a data pipeline for a real-time recommendation system. The pipeline must handle high velocity streams and ensure data quality. Which three components should be included in the pipeline? (Select THREE).

Accepted Answer

A stream processing engine like Apache Kafka Streams. Option A (a stream processing engine like Apache Kafka Streams) is correct because a real-time recommendation pipeline requires continuous, low-latency processing of high-velocity events, and Kafka Streams provides exactly that with stateful stream operations on Kafka topics. Option B (a data validation step to check schema compliance) is correct because the scenario explicitly demands ensuring data quality, and schema validation catches malformed or non-conforming records before they contaminate downstream recommendations. Option E (a message queue for buffering) is correct because high-velocity streams need a durable buffer to absorb bursts and decouple producers from consumers, preventing data loss and backpressure; Kafka itself is commonly used as this queue. Option C (a data warehouse for historical analysis) is not required for the real-time path since it serves analytical/OLAP workloads rather than low-latency recommendations. Option D (a batch processing framework like Apache Spark) is not appropriate here because batch processing introduces latency and does not satisfy the real-time streaming requirement.

Answer

A data warehouse for historical analysis

Answer

A batch processing framework like Apache Spark

Question 9

Refer to the exhibit. A data engineer notices that the batch processing step is taking too long and causing delays. Which change would most likely reduce the latency?

Accepted Answer

Move feature engineering to the stream processing step in Flink. Moving feature engineering from the batch processing step to the stream processing step in Flink reduces the volume of data that the batch job must process and can eliminate redundant computations. By computing features in real-time as data streams in, the batch job can focus on model training or aggregation, thereby reducing overall latency. This architectural shift is a common optimization in hybrid batch-stream pipelines.

Answer

Increase the parallelism of the Spark job

Answer

Replace Apache Flink with Apache Storm for stream processing

Answer

Change the output format from Parquet to CSV

Question 10

A data scientist is preparing a dataset for a supervised learning model. The dataset contains missing values in 15% of the rows for a numeric feature. Which preprocessing technique should be applied to minimize bias?

Accepted Answer

Impute missing values with the mean of the feature.. Imputing missing values with the mean of the feature is the correct approach because it preserves the overall distribution of the numeric feature and avoids introducing systematic bias when only 15% of rows are affected. Removing rows would reduce sample size and potentially introduce selection bias, while encoding missing values as a separate category is inappropriate for numeric features in regression or classification tasks. Mean imputation is a standard, low-bias technique for MCAR (Missing Completely At Random) data in supervised learning pipelines.

Answer

Remove all rows with missing values.

Answer

Encode the missing values as a separate category.

Answer

Use a model that handles missing values natively.

Question 11

An AI team notices that a model's F1 score on the validation set is 0.95, but on the test set it drops to 0.72. Which course of action is most appropriate?

Accepted Answer

Apply regularization.. The F1 score dropping from 0.95 on the validation set to 0.72 on the test set is a classic sign of overfitting, where the model has memorized the training/validation data but fails to generalize to unseen test data. Applying regularization (e.g., L1/L2 weight decay, dropout) is the most appropriate course of action because it penalizes overly complex models, reduces variance, and improves generalization without requiring more data or changing the model architecture.

Answer

Reduce the training dataset size.

Answer

Adjust the train/test split to be more representative.

Answer

Increase model complexity.

Question 12

A model trained on a dataset with imbalanced classes achieves 98% accuracy but only 50% recall for the minority class. Which technique should be applied first to address the imbalance?

Accepted Answer

Apply cost-sensitive learning. Cost-sensitive learning directly modifies the model's loss function to penalize misclassifications of the minority class more heavily than those of the majority class. This approach addresses the root cause of the imbalance—the model's bias toward the majority class—without altering the dataset distribution, making it the most immediate and effective first step.

Answer

Reduce the majority class size

Answer

Use SMOTE to generate synthetic samples

Answer

Collect more data for the minority class

Question 13

A data scientist is evaluating a logistic regression model for binary classification on highly imbalanced data. Which TWO metrics are most appropriate to assess model performance? (Choose TWO.)

Accepted Answer

Recall. Recall (Option B) is correct because in highly imbalanced binary classification, it measures the proportion of actual positives correctly identified (TP/(TP+FN)), which is critical when missing the minority class is costly. Precision (Option C) is correct because it measures the proportion of predicted positives that are truly positive (TP/(TP+FP)), helping assess the reliability of positive predictions when false positives matter. Accuracy (Option A) is not appropriate because with severe class imbalance a model can achieve high accuracy by predicting only the majority class while performing poorly on the minority class. Mean squared error (Option D) is a regression loss and is not a standard classification performance metric for logistic regression class labels. F1 score (Option E) is a useful combined measure of precision and recall, but it is not one of the two options marked correct here.

Answer

Accuracy

Answer

Mean squared error (MSE)

Answer

F1 score

Question 14

A team is using k-fold cross-validation to evaluate a model. They observe high variance in performance scores across folds. Which TWO actions are most likely to reduce this variance? (Choose TWO.)

Accepted Answer

Increase the number of folds. Increasing the number of folds (A) is correct because with more folds each training set contains a larger fraction of the data and each validation set is smaller, so the individual fold estimates are based on more similar training distributions, which reduces the variance of the per-fold scores and of the overall mean estimate. Using stratified cross-validation (B) is correct because it preserves the class (or target) proportions in every fold, preventing folds from having unrepresentative label distributions that cause erratic scores, which is especially important with imbalanced data. Decreasing the number of folds (C) is wrong because fewer folds means smaller training sets and larger, more variable validation sets, which typically increases variance. Shuffling data before splitting (D) is wrong because it only randomizes order and does not by itself control the fold-to-fold variability; without stratification it can even worsen imbalance across folds. Using a more complex model (E) is wrong because higher model complexity generally increases variance in performance estimates rather than reducing it.

Answer

Decrease the number of folds

Answer

Shuffle data before splitting

Answer

Use a more complex model

Question 15

A real-time recommendation system requires low latency. Which data storage strategy is best for serving user profiles and item embeddings?

Accepted Answer

In-memory key-value store (e.g., Redis). An in-memory key-value store like Redis is ideal for serving user profiles and item embeddings in a real-time recommendation system because it provides sub-millisecond read/write latency by keeping data in RAM. This directly meets the low-latency requirement for fetching embeddings and profiles on each request, without the overhead of disk I/O or complex query processing.

Answer

Time-series database (e.g., InfluxDB)

Answer

Relational database with joins

Answer

Data lake in object storage (e.g., Amazon S3)