I85PREP
⌘ K
Alteryx Machine Learning study materials
Alteryx Machine Learning study resources

Alteryx Machine Learning practice questions

One question at a time. Build your Alteryx Machine Learning understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Machine Learning Fundamentals and Problem Framing

A retail analyst wants to predict whether a customer will churn next month using historical account activity. Which problem type best describes this task?

  1. Time-series forecasting
  2. Clustering
  3. Binary classification Correct answer.
  4. Regression

Explanation

The target is a two-valued label (churn or no churn), which makes it a binary classification problem. Regression predicts continuous values, clustering is unsupervised, and forecasting projects future numeric values.

Question 2Machine Learning Fundamentals and Problem Framing

In supervised machine learning, what is the role of the target variable?

  1. It is the feature the model weights most heavily
  2. It is the identifier that uniquely labels each row
  3. It is the column removed to prevent data leakage
  4. It is the outcome the model is trained to predict Correct answer.

Explanation

The target variable is the known outcome in the training data that the model learns to predict. Features are the inputs used to make that prediction.

Question 3Machine Learning Fundamentals and Problem Framing

Which statement describes unsupervised learning?

  1. It always produces a numeric prediction
  2. It requires labeled outcomes for every row
  3. It models relationships without a target column Correct answer.
  4. It can only be applied to text data

Explanation

Unsupervised learning finds structure in data without a predefined target label. Clustering and dimensionality reduction are common examples.

Question 4Machine Learning Fundamentals and Problem Framing

A dataset contains 9,500 legitimate transactions and 500 fraudulent ones. Which issue should the analyst address first?

  1. Heteroscedasticity
  2. Severe class imbalance Correct answer.
  3. Multicollinearity
  4. Autocorrelation

Explanation

The minority class is only about 5% of the data, which is class imbalance and can bias models toward the majority class. Multicollinearity, heteroscedasticity, and autocorrelation are regression diagnostics unrelated to this ratio.

Question 5Machine Learning Fundamentals and Problem Framing

Which pair correctly matches a task to its typical evaluation metric?

  1. Regression and F1 score
  2. Binary classification and ROC AUC Correct answer.
  3. Dimensionality reduction and precision
  4. Clustering and mean absolute error

Explanation

ROC AUC measures how well a classifier separates classes across thresholds, so it suits binary classification. F1 is a classification metric, MAE is a regression metric, and precision is a classification metric.

Question 6Machine Learning Fundamentals and Problem Framing

An analyst splits data into training and validation sets. What is the primary purpose of the validation set?

  1. To increase the number of features
  2. To store rows excluded from modeling
  3. To fit the model parameters
  4. To estimate performance on unseen data Correct answer.

Explanation

The validation set provides an unbiased estimate of how the model performs on data it was not trained on. Model parameters are fit on the training set.

Question 7Machine Learning Fundamentals and Problem Framing

A model scores 0.98 accuracy on training data but 0.62 on holdout data. What does this pattern indicate?

  1. Class imbalance
  2. Data leakage
  3. Overfitting Correct answer.
  4. Underfitting

Explanation

A large gap between high training performance and much lower holdout performance is the signature of overfitting. The model has memorized training patterns that do not generalize.

Question 8Machine Learning Fundamentals and Problem Framing

Which practice helps prevent data leakage when preparing features?

  1. Duplicate rows across training and holdout sets
  2. Include the target column as an input feature
  3. Fit imputation and scaling on the training data only Correct answer.
  4. Compute feature statistics on the full dataset before splitting

Explanation

Leakage occurs when information from the holdout set influences training, so transformations such as imputation and scaling must be fit on training data only. Computing statistics on the full dataset or including the target leaks outcome information.

Question 9Machine Learning Fundamentals and Problem Framing

Which statement best describes a feature in a machine learning model?

  1. The algorithm that fits the model
  2. The predicted label produced by the model
  3. The error between prediction and truth
  4. An input variable used to make predictions Correct answer.

Explanation

Features are the input variables the model uses to predict the target. The predicted label is the model output, and error is a performance measure.

Question 10Machine Learning Fundamentals and Problem Framing

A team wants to group customers into segments without predefined segment labels. Which approach fits this goal?

  1. k-means clustering Correct answer.
  2. Logistic regression
  3. Gradient boosting for classification
  4. Linear regression

Explanation

Segmentation without labels is an unsupervised clustering task, and k-means is a clustering algorithm. The other options are supervised methods that require a target.

Question 11Machine Learning Fundamentals and Problem Framing

Why is a baseline model useful at the start of a project?

  1. It provides a reference point to judge improvement Correct answer.
  2. It guarantees the best achievable accuracy
  3. It eliminates class imbalance
  4. It removes the need for a holdout set

Explanation

A baseline gives a simple reference score so the team can tell whether a more complex model adds real value. It does not guarantee optimal performance or remove the need for validation.

Question 12Machine Learning Fundamentals and Problem Framing

What distinguishes a classification task from a regression task?

  1. Classification requires text; regression requires numeric inputs
  2. Classification is unsupervised; regression is supervised
  3. Classification predicts a category; regression predicts a number Correct answer.
  4. Classification uses no training data; regression does

Explanation

Classification predicts discrete class labels, while regression predicts continuous numeric values. Both are supervised and both can use numeric or text-derived inputs.

Question 13Machine Learning Fundamentals and Problem Framing

A model predicts loan default with high recall but low precision. What is the practical consequence?

  1. The model cannot be evaluated
  2. Many actual defaults are missed
  3. The threshold is set too high
  4. Many good loans are incorrectly flagged as defaults Correct answer.

Explanation

High recall with low precision means the model catches most defaults but also flags many non-defaults, producing false positives. Raising the threshold typically reduces false positives but can lower recall.

Question 14Machine Learning Fundamentals and Problem Framing

Which scenario is a clear example of a regression problem?

  1. Labeling an image as cat or dog
  2. Predicting next month's sales revenue Correct answer.
  3. Assigning customers to unnamed clusters
  4. Predicting whether an email is spam

Explanation

Sales revenue is a continuous numeric quantity, so predicting it is regression. Spam detection and image labeling are classification, and unnamed grouping is clustering.

Question 15Machine Learning Fundamentals and Problem Framing

What is the main risk of using too many irrelevant features in a model?

  1. The training set grows too large
  2. The model may fit noise and generalize poorly Correct answer.
  3. The target variable becomes continuous
  4. The evaluation metric becomes invalid

Explanation

Irrelevant features add noise and can cause the model to overfit, reducing generalization. Feature selection or dimensionality reduction helps mitigate this.

Question 16Machine Learning Fundamentals and Problem Framing

Why should a holdout or test set be kept separate until final evaluation?

  1. To avoid encoding categorical variables
  2. To provide an unbiased estimate of generalization Correct answer.
  3. To increase the number of training rows
  4. To speed up model training

Explanation

Keeping the test set untouched until the end ensures the final performance estimate reflects unseen data. Repeatedly tuning on it would bias the estimate.

Question 17Machine Learning Fundamentals and Problem Framing

An analyst builds a model to predict a customer's exact age in years from purchase history. Which framing is most appropriate?

  1. Anomaly detection, because exact age is rare
  2. Clustering, because customers are grouped
  3. Classification, because age is an integer
  4. Regression, because age is a continuous quantity Correct answer.

Explanation

Age is a continuous numeric quantity, so predicting it is regression even if recorded as whole years. Classification applies to discrete categories, not ordered numeric values.

Question 18Machine Learning Fundamentals and Problem Framing

Which metric is most appropriate when the cost of a false negative is much higher than a false positive?

  1. Accuracy
  2. Mean squared error
  3. Recall Correct answer.
  4. Specificity

Explanation

Recall measures the share of actual positives that are correctly identified, so it directly reflects missed positives. When false negatives are costly, high recall is prioritized.

Question 19Machine Learning Fundamentals and Problem Framing

What does a confusion matrix summarize for a binary classifier?

  1. The importance of each feature
  2. Counts of predicted versus actual classes Correct answer.
  3. The distribution of a numeric target
  4. The correlation between two features

Explanation

A confusion matrix tabulates actual versus predicted class counts, enabling computation of precision, recall, and related metrics. It does not describe feature correlation or target distribution.

Question 20Machine Learning Fundamentals and Problem Framing

A model performs well in development but poorly when deployed on new data from a different region. What is the most likely cause?

  1. The model uses too few features
  2. The deployment data has more columns
  3. The evaluation metric was regression-based
  4. The model overfits the training region Correct answer.

Explanation

When deployment data differs from training data, a model that fit region-specific patterns may not generalize, a form of distribution shift tied to overfitting. The issue is representativeness, not column count or metric choice.

Question 21Machine Learning Fundamentals and Problem Framing

Which activity belongs to the problem-framing stage of a machine learning project?

  1. Deploying the model to production
  2. Encoding categorical variables
  3. Tuning hyperparameters
  4. Defining the target and success metric Correct answer.

Explanation

Problem framing establishes what is being predicted and how success will be measured before modeling begins. Tuning, encoding, and deployment occur in later stages.

Question 22Machine Learning Fundamentals and Problem Framing

Why is it important to check that the training data is representative of the population the model will score?

  1. To remove the need for a validation set
  2. To ensure the model generalizes to real cases Correct answer.
  3. To guarantee perfect accuracy
  4. To reduce the number of features needed

Explanation

A model learns patterns from its training data, so unrepresentative samples lead to poor performance on the real population. Representativeness supports generalization but does not guarantee perfect accuracy.

Question 23Machine Learning Fundamentals and Problem Framing

What is the purpose of feature engineering?

  1. To create informative input variables from raw data Correct answer.
  2. To store predictions for deployment
  3. To select the learning algorithm
  4. To measure model error on holdout data

Explanation

Feature engineering transforms raw data into inputs that better expose patterns to the model, such as ratios or date parts. Algorithm selection and error measurement are separate steps.

Question 24Machine Learning Fundamentals and Problem Framing

A dataset has a numeric target with a few extreme outliers. Which concern is most relevant for model evaluation?

  1. Clustering must replace regression
  2. The target should be converted to a class label
  3. Accuracy becomes the only valid metric
  4. Metrics like mean squared error may be dominated by outliers Correct answer.

Explanation

Squared-error metrics penalize large deviations heavily, so outliers can dominate the score. Robust metrics or outlier handling may be warranted, but the target type does not change.

Question 25Machine Learning Fundamentals and Problem Framing

What does it mean for a machine learning model to generalize?

  1. It memorizes the training rows exactly
  2. It uses the fewest possible features
  3. It trains in the shortest time
  4. It performs well on data it was not trained on Correct answer.

Explanation

Generalization is the ability to make accurate predictions on new, unseen data. Memorizing training rows is overfitting, and feature count or training speed do not define generalization.

Question 26Machine Learning Fundamentals and Problem Framing

In a machine learning project, which activity is part of defining the problem?

  1. Deploying the model to production
  2. Tuning the learning rate
  3. Cleaning the dataset
  4. Selecting the evaluation metric Correct answer.

Explanation

Problem definition includes specifying the goal, the target variable, and how success will be measured, which requires choosing an evaluation metric. The other activities occur in later stages such as data preparation, model training, and deployment.

Question 27Machine Learning Fundamentals and Problem Framing

A hospital wants to predict whether a patient will be readmitted within 30 days. Which type of machine learning task is this?

  1. Regression
  2. Classification Correct answer.
  3. Dimensionality reduction
  4. Clustering

Explanation

The target is a binary outcome (readmitted or not), so the task is classification. Regression predicts a continuous quantity, while clustering and dimensionality reduction are unsupervised techniques.

Question 28Machine Learning Fundamentals and Problem Framing

A model predicts the probability that a customer will churn. Which evaluation metric is most appropriate when the business wants to rank customers by their likelihood to churn?

  1. Mean squared error
  2. ROC AUC Correct answer.
  3. Accuracy
  4. R-squared

Explanation

ROC AUC measures how well the model ranks positive instances above negative ones, which aligns with the goal of ranking customers by churn probability. Accuracy requires a fixed threshold, MSE and R-squared are regression metrics.

Question 29Machine Learning Fundamentals and Problem Framing

A team is building a model to predict whether a loan applicant will default. They have a dataset with 10,000 applicants, of whom only 200 defaulted. Which issue is most important to address during problem framing?

  1. The classes are highly imbalanced. Correct answer.
  2. The dataset is too large for standard algorithms.
  3. The target variable is continuous.
  4. The features are all categorical.

Explanation

With only 2% of applicants defaulting, the classes are highly imbalanced, which can cause a model to ignore the minority class. This should be addressed through resampling, class weights, or appropriate metrics. Dataset size, target type, and feature types are not the primary concern here.

Question 30Machine Learning Fundamentals and Problem Framing

A data scientist is asked to predict next quarter's sales for each of 50 stores. She has historical quarterly sales for each store and several economic indicators. Which consideration is most critical when framing this problem?

  1. Whether to reduce the number of stores to simplify the model
  2. How to encode the store identifier as a feature
  3. Whether to use a classification algorithm instead
  4. How to handle the time series nature of the data to avoid leakage Correct answer.

Explanation

Because the data are sequential over time, using future information to predict the past would cause leakage; proper time-based splitting and feature timing must be addressed. The target is continuous, so classification is not appropriate; store ID encoding and reducing stores are secondary concerns.

Question 31Data Preparation and Feature Engineering

In Alteryx Machine Learning, what is the primary purpose of the data preparation stage before model training?

  1. To compute final evaluation metrics on the test set
  2. To clean, transform, and shape raw data into a modeling-ready table Correct answer.
  3. To select the algorithm with the highest accuracy automatically
  4. To deploy the trained model to a production scoring endpoint

Explanation

Data preparation focuses on cleaning, transforming, and shaping raw inputs so the modeling step receives a consistent, usable table. Evaluation metrics and deployment occur after a model is trained, and algorithm selection is a separate modeling concern.

Question 32Data Preparation and Feature Engineering

A column in an Alteryx Machine Learning dataset contains the values 'yes', 'no', and 'unknown'. Which treatment is most appropriate before modeling?

  1. Encode it as a categorical variable with distinct levels Correct answer.
  2. Convert each value to its character count
  3. Treat it as a continuous numeric field
  4. Drop the column because it has three values

Explanation

Non-numeric fields with a small set of distinct values are categorical and should be encoded as such so the model can use them. Converting to character counts or treating them as numbers would fabricate meaningless numeric relationships.

Question 33Data Preparation and Feature Engineering

What does one-hot encoding produce for a categorical column?

  1. A normalized continuous version of the original column
  2. One binary indicator column per distinct category level Correct answer.
  3. A single numeric column with arbitrary integer codes
  4. A text column containing concatenated category names

Explanation

One-hot encoding creates a separate binary column for each category level, marking presence with 1 and absence with 0. This avoids implying an order among categories.

Question 34Data Preparation and Feature Engineering

Why is imputation used when a numeric feature contains missing values?

  1. To convert the numeric feature into a categorical one
  2. To replace missing entries with a plausible value so rows can still be used Correct answer.
  3. To delete every row that contains any missing entry
  4. To increase the number of rows in the dataset

Explanation

Imputation fills missing entries with an estimated value, such as the mean or median, so those records remain usable. It does not create new rows or change the fundamental type of the feature.

Question 35Data Preparation and Feature Engineering

A numeric feature has a highly right-skewed distribution with extreme outliers. Which transformation is commonly applied to reduce that skew?

  1. Applying a logarithmic transformation Correct answer.
  2. Multiplying every value by a constant
  3. Sorting the values in ascending order
  4. Rounding values to the nearest integer

Explanation

A log transformation compresses large values more than small ones, reducing right skew and the influence of extreme outliers. Scaling by a constant or reordering values does not change the shape of the distribution.

Question 36Data Preparation and Feature Engineering

What is the effect of standardizing a numeric feature to zero mean and unit variance?

  1. It converts the feature into a categorical variable
  2. It removes all outliers from the feature
  3. It puts features on a comparable scale for modeling Correct answer.
  4. It changes the distribution into a uniform one

Explanation

Standardization rescales a feature so it has mean zero and standard deviation one, placing features with different units on a comparable scale. It does not alter the underlying distribution shape or remove outliers.

Question 37Data Preparation and Feature Engineering

When splitting data into training and validation sets, why should the split be performed before fitting imputation or scaling parameters?

  1. To prevent information from the validation set leaking into preprocessing Correct answer.
  2. To avoid the need for any evaluation metrics
  3. To guarantee the model achieves perfect accuracy
  4. To make the training set larger than the full dataset

Explanation

Fitting preprocessing on the full dataset lets statistics from validation data influence training, a form of data leakage. Parameters should be learned on the training portion and then applied to validation data.

Question 38Data Preparation and Feature Engineering

A dataset column records the date each customer account was opened. Which engineered feature is most likely to be useful for predicting churn?

  1. The raw date string unchanged
  2. Account age in days as of the reference date Correct answer.
  3. The number of characters in the date string
  4. The alphabetical order of the month name

Explanation

Deriving account age captures the elapsed time since opening, which is directly meaningful for churn. The raw date string and character counts carry no usable numeric relationship.

Question 39Data Preparation and Feature Engineering

What is the main risk of target leakage in feature engineering?

  1. The dataset requires more storage space
  2. The model appears accurate in testing but fails in production Correct answer.
  3. The categorical features cannot be encoded
  4. The model trains more slowly than expected

Explanation

Target leakage occurs when a feature contains information about the outcome that would not be available at prediction time, inflating test performance. Such models typically underperform when deployed on genuinely new data.

Question 40Data Preparation and Feature Engineering

In Alteryx Machine Learning, what does the target variable represent?

  1. The row number assigned during preparation
  2. The column the model is trained to predict Correct answer.
  3. The feature with the highest variance
  4. The identifier column used to join tables

Explanation

The target variable is the outcome the model learns to predict from the input features. Identifiers and row numbers are not outcomes and should typically be excluded from modeling.

Question 41Data Preparation and Feature Engineering

Which column should generally be excluded from the feature set before training?

  1. A unique record identifier column Correct answer.
  2. A feature correlated with the target
  3. A numeric feature with some missing values
  4. A categorical feature with many levels

Explanation

Unique identifiers carry no generalizable signal and can cause the model to memorize records. Numeric and categorical features, even imperfect ones, can still contribute predictive information.

Question 42Data Preparation and Feature Engineering

What is the purpose of binning a continuous numeric feature?

  1. To remove duplicate rows from the dataset
  2. To group values into discrete ranges or buckets Correct answer.
  3. To increase the precision of the original values
  4. To convert the target into a feature

Explanation

Binning groups continuous values into discrete ranges, which can simplify relationships and reduce sensitivity to small variations. It does not affect duplicates, the target, or numeric precision.

Question 43Data Preparation and Feature Engineering

A categorical feature has hundreds of distinct levels, many appearing only a few times. Which approach best reduces dimensionality while retaining signal?

  1. Group rare levels into a single 'other' category Correct answer.
  2. One-hot encode every level without change
  3. Delete the feature entirely
  4. Convert the feature to a numeric row index

Explanation

Grouping infrequent levels into an 'other' bucket reduces cardinality and prevents sparse, unstable indicator columns. Encoding every level or deleting the feature loses or bloats information unnecessarily.

Question 44Data Preparation and Feature Engineering

Which statement correctly describes the difference between normalization (min-max scaling) and standardization (z-score scaling)?

  1. Normalization applies only to text, while standardization applies only to dates
  2. Normalization removes missing values, while standardization imputes them
  3. Normalization maps values to a fixed range, while standardization centers on mean and scales by standard deviation Correct answer.
  4. Normalization and standardization are identical operations with different names

Explanation

Min-max normalization rescales values into a bounded range such as 0 to 1, while standardization produces zero mean and unit variance. They are distinct rescaling methods, not interchangeable or type-specific operations.

Question 45Data Preparation and Feature Engineering

What does feature selection aim to accomplish?

  1. Increase the number of rows available for training
  2. Retain the most informative features and remove redundant or irrelevant ones Correct answer.
  3. Replace the target variable with a proxy
  4. Generate entirely new columns from existing ones

Explanation

Feature selection identifies and keeps the most useful predictors while removing weak or redundant ones, which can improve model performance and interpretability. Creating new columns is feature engineering, not selection.

Question 46Data Preparation and Feature Engineering

A feature records income in dollars with a few extremely large values. Which preprocessing step most directly reduces the influence of those extreme values on distance-based models?

  1. Removing the target column
  2. Applying a log or robust scaling transform Correct answer.
  3. Adding a random noise column
  4. Duplicating the extreme rows

Explanation

Log or robust scaling compresses or rescales extreme values so they do not dominate distance calculations. Adding noise, duplicating rows, or removing the target does not address the outlier influence.

Question 47Data Preparation and Feature Engineering

Why is it important to handle missing values consistently between training data and new scoring data?

  1. Because imputation parameters must be applied the same way to avoid inconsistent inputs Correct answer.
  2. Because missing values are not allowed in any dataset
  3. Because the target variable must also be imputed
  4. Because scoring data must contain identical row counts

Explanation

Imputation parameters learned on training data must be reused when scoring so features are treated consistently. Inconsistent handling would shift the input distribution relative to what the model learned.

Question 48Data Preparation and Feature Engineering

What is the purpose of creating interaction features between two numeric variables?

  1. To remove both original variables from the dataset
  2. To guarantee the model ignores the original features
  3. To convert numeric variables into text
  4. To capture combined effects that neither variable shows alone Correct answer.

Explanation

Interaction features, such as a product or ratio, represent combined effects that individual variables may not reveal. They supplement rather than replace the original variables and remain numeric.

Question 49Data Preparation and Feature Engineering

A date field is decomposed into year, month, day-of-week, and day-of-year. What is the main benefit of this decomposition?

  1. It guarantees the model will not overfit
  2. It removes the need for a target variable
  3. It exposes cyclical and seasonal patterns the model can use Correct answer.
  4. It reduces the dataset's storage footprint

Explanation

Splitting a date into components exposes seasonal and cyclical patterns, such as weekly or monthly effects, that a raw timestamp does not present directly. It does not eliminate the target or guarantee against overfitting.

Question 50Data Preparation and Feature Engineering

Which situation best indicates that a feature should be removed because it is constant across all records?

  1. The feature contains a small number of missing values
  2. The feature is numeric rather than categorical
  3. The feature holds the same value for every row Correct answer.
  4. The feature has a different value in every row

Explanation

A feature with the same value in every row has zero variance and cannot help distinguish outcomes, so it provides no predictive value. Features with missing values or numeric types can still be informative.