Skip to content
Zurück zu den Lernmaterialien

50 Free AI/ML Certification Practice Questions

29. Juli 2026~17 min read

Preparing for AI/ML certification (AWS ML Specialty, Azure AI Engineer, or GCP ML Engineer)? These 50 free practice questions cover core ML concepts and cloud ML services.


Section 1: Machine Learning Fundamentals (Questions 1–20)

Question 1

Which type of ML is used when the target variable is continuous (e.g., predicting house prices)?

a) Classification
b) Regression
c) Clustering
d) Association

Show Answer

Answer: b) Regression

Explanation: Regression predicts continuous numerical values. Linear regression, polynomial regression, and random forest regression are examples. Classification predicts discrete categories.

Question 2

Which ML algorithm is best for classifying emails as spam or not spam?

a) Linear Regression
b) Logistic Regression
c) K-Means
d) Apriori

Show Answer

Answer: b) Logistic Regression

Explanation: Despite its name, logistic regression is used for binary classification problems. It outputs probabilities between 0 and 1, making it ideal for spam detection.

Question 3

What is overfitting in machine learning?

a) Model performs well on training data but poorly on new data
b) Model performs poorly on both training and test data
c) Model performs well on test data but poorly on training data
d) Model is too simple

Show Answer

Answer: a) Model performs well on training data but poorly on new data

Explanation: Overfitting occurs when a model learns noise and details in training data too well, failing to generalize to unseen data. Solutions include regularization, cross-validation, and more data.

Question 4

What is the purpose of splitting data into training and test sets?

a) To make the model train faster
b) To evaluate model performance on unseen data
c) To reduce memory usage
d) To increase accuracy

Show Answer

Answer: b) To evaluate model performance on unseen data

Explanation: Splitting data allows us to evaluate how well the model generalizes to new, unseen data. The training set is used to learn patterns; the test set evaluates performance.

Question 5

What is cross-validation?

a) Training multiple models simultaneously
b) Splitting data into k-folds and training/evaluating k times
c) Validating data before training
d) Cross-referencing predictions

Show Answer

Answer: b) Splitting data into k-folds and training/evaluating k times

Explanation: K-fold cross-validation divides data into k equal folds. The model trains on k-1 folds and validates on the remaining fold, repeated k times. This gives a more robust performance estimate.

Question 6

Which metric is appropriate for evaluating a binary classification model?

a) Mean Squared Error (MSE)
b) R-squared
c) Accuracy, Precision, Recall, F1-Score
d) Mean Absolute Error (MAE)

Show Answer

Answer: c) Accuracy, Precision, Recall, F1-Score

Explanation: Classification metrics include accuracy (overall correctness), precision (positive predictions that are correct), recall (actual positives correctly identified), and F1-score (harmonic mean of precision and recall).

Question 7

Which metric is appropriate for evaluating a regression model?

a) Accuracy
b) Precision
c) Mean Squared Error (MSE)
d) F1-Score

Show Answer

Answer: c) Mean Squared Error (MSE)

Explanation: Regression metrics include MSE (average squared difference between predictions and actual values), RMSE, MAE, and R-squared.

Question 8

What is the purpose of feature scaling (normalization/standardization)?

a) Reducing the number of features
b) Bringing all features to a similar scale for better model performance
c) Creating new features
d) Removing outliers

Show Answer

Answer: b) Bringing all features to a similar scale for better model performance

Explanation: Feature scaling ensures all features contribute equally, preventing features with larger ranges from dominating. It's essential for algorithms like SVM, KNN, and neural networks.

Question 9

What is the difference between bagging and boosting?

a) Bagging trains models in parallel; boosting trains sequentially
b) Bagging trains sequentially; boosting trains in parallel
c) They are the same
d) Bagging is for regression; boosting is for classification

Show Answer

Answer: a) Bagging trains models in parallel; boosting trains sequentially

Explanation: Bagging (e.g., Random Forest) trains multiple models independently in parallel and averages predictions. Boosting (e.g., XGBoost) trains models sequentially, each correcting the errors of the previous.

Question 10

Which algorithm is commonly used for unsupervised clustering?

a) Linear Regression
b) Logistic Regression
c) K-Means
d) Decision Tree

Show Answer

Answer: c) K-Means

Explanation: K-Means is a popular unsupervised clustering algorithm that partitions data into k clusters based on similarity. It's used for customer segmentation, image compression, and anomaly detection.

Question 11

What is the purpose of a confusion matrix?

a) Visualizing data distribution
b) Showing true vs predicted classifications
c) Calculating feature importance
d) Reducing dimensionality

Show Answer

Answer: b) Showing true vs predicted classifications

Explanation: A confusion matrix shows counts of True Positives, True Negatives, False Positives, and False Negatives, enabling calculation of accuracy, precision, recall, and F1-score.

Question 12

What is the bias-variance tradeoff?

a) Tradeoff between model accuracy and training time
b) Tradeoff between underfitting (bias) and overfitting (variance)
c) Tradeoff between model size and performance
d) Tradeoff between data size and compute

Show Answer

Answer: b) Tradeoff between underfitting (bias) and overfitting (variance)

Explanation: High bias (underfitting) means the model is too simple to capture patterns. High variance (overfitting) means the model is too complex and captures noise. The goal is to find the sweet spot.

Question 13

What is regularization in machine learning?

a) Adding more training data
b) Adding a penalty to the loss function to prevent overfitting
c) Removing features
d) Increasing model complexity

Show Answer

Answer: b) Adding a penalty to the loss function to prevent overfitting

Explanation: Regularization (L1 Lasso, L2 Ridge) adds a penalty term to the loss function to constrain model coefficients, reducing overfitting by discouraging complex models.

Question 14

What is the difference between L1 and L2 regularization?

a) L1 can shrink coefficients to zero (feature selection); L2 shrinks coefficients but not to zero
b) L2 can shrink coefficients to zero; L1 can't
c) There is no difference
d) L1 is for classification; L2 is for regression

Show Answer

Answer: a) L1 can shrink coefficients to zero (feature selection); L2 shrinks coefficients but not to zero

Explanation: L1 (Lasso) adds the absolute value of coefficients as a penalty, which can zero out less important features. L2 (Ridge) adds the squared magnitude, shrinking coefficients but never to zero.

Question 15

What is a hyperparameter?

a) A parameter learned from training data
b) A parameter set before training that controls the learning process
c) A feature in the dataset
d) A model output

Show Answer

Answer: b) A parameter set before training that controls the learning process

Explanation: Hyperparameters are configuration settings set before training (learning rate, number of trees, regularization strength). Model parameters (weights, biases) are learned during training.

Question 16

Which technique is used to find the best hyperparameters?

a) Gradient descent
b) Grid search or random search
c) Backpropagation
d) Forward propagation

Show Answer

Answer: b) Grid search or random search

Explanation: Grid search exhaustively tests all combinations of hyperparameters. Random search randomly samples hyperparameter combinations and is often more efficient for high-dimensional spaces.

Question 17

What is the purpose of a validation set?

a) Training the final model
b) Tuning hyperparameters and model selection
c) Testing on unseen data
d) Data preprocessing

Show Answer

Answer: b) Tuning hyperparameters and model selection

Explanation: The validation set (held out from training data) is used to tune hyperparameters and compare model performance. The test set is only used for final evaluation.

Question 18

What is an ROC curve?

a) A plot of true positive rate vs false positive rate at various thresholds
b) A plot of accuracy vs model complexity
c) A plot of training vs test error
d) A plot of feature importance

Show Answer

Answer: a) A plot of true positive rate vs false positive rate at various thresholds

Explanation: The ROC curve shows the trade-off between true positive rate (sensitivity) and false positive rate (1-specificity) across different classification thresholds. AUC is the area under this curve.

Question 19

What does AUC (Area Under the Curve) measure?

a) Model accuracy
b) Model's ability to distinguish between classes
c) Model training time
d) Number of features needed

Show Answer

Answer: b) Model's ability to distinguish between classes

Explanation: AUC measures the model's ability to separate positive and negative classes. AUC = 1 is perfect; AUC = 0.5 is random guessing.

Question 20

Which of the following is an ensemble method?

a) Linear Regression
b) Random Forest
c) K-Means
d) PCA

Show Answer

Answer: b) Random Forest

Explanation: Random Forest is an ensemble method that combines multiple decision trees using bagging. It reduces overfitting and improves accuracy compared to single decision trees.


Section 2: Cloud ML Services (Questions 21–35)

Question 21

Which AWS SageMaker component is used for training ML models at scale?

a) SageMaker Studio
b) SageMaker Training Jobs
c) SageMaker Ground Truth
d) SageMaker Neo

Show Answer

Answer: b) SageMaker Training Jobs

Explanation: SageMaker Training Jobs launch managed ML training infrastructure with automatic scaling, checkpointing, and distributed training support.

Question 22

Which AWS service provides pre-trained AI for image analysis?

a) Amazon Rekognition
b) Amazon Comprehend
c) Amazon Polly
d) Amazon Lex

Show Answer

Answer: a) Amazon Rekognition

Explanation: Amazon Rekognition provides pre-trained AI for image and video analysis: object detection, facial recognition, content moderation, and text detection.

Question 23

Which Azure AI service provides natural language understanding?

a) Azure Computer Vision
b) Azure Language Service
c) Azure Speech
d) Azure Search

Show Answer

Answer: b) Azure Language Service

Explanation: Azure Language Service provides pre-built NLP capabilities: sentiment analysis, key phrase extraction, entity recognition, and language understanding (LUIS).

Question 24

Which GCP service provides a unified ML platform for building and deploying models?

a) Cloud AutoML
b) Vertex AI
c) AI Platform
d) Cloud ML Engine

Show Answer

Answer: b) Vertex AI

Explanation: Vertex AI is Google Cloud's unified ML platform, integrating AutoML, custom training, model deployment, and MLOps tools into a single service.

Question 25

What is the purpose of SageMaker Ground Truth?

a) Model training
b) Data labeling
c) Model deployment
d) Feature engineering

Show Answer

Answer: b) Data labeling

Explanation: SageMaker Ground Truth helps create labeled datasets using human annotators (via mechanical turk or private workforce) and automated labeling.

Question 26

Which AWS service translates text between languages?

a) Amazon Translate
b) Amazon Transcribe
c) Amazon Polly
d) Amazon Lex

Show Answer

Answer: a) Amazon Translate

Explanation: Amazon Translate is a neural machine translation service that translates text between supported languages. Transcribe is for speech-to-text, Polly for text-to-speech.

Question 27

Which Azure Cognitive Service provides speech-to-text?

a) Azure Speech-to-Text
b) Azure Text-to-Speech
c) Azure Translator
d) Azure Language

Show Answer

Answer: a) Azure Speech-to-Text

Explanation: Azure Speech-to-Text (part of Azure Speech Services) converts audio streams to text in real-time or batch, with customization for domain-specific vocabulary.

Question 28

Which GCP service provides pre-trained vision AI APIs?

a) Cloud Vision API
b) Cloud Translation API
c) Cloud Natural Language API
d) Cloud Speech-to-Text

Show Answer

Answer: a) Cloud Vision API

Explanation: Cloud Vision API provides pre-trained AI models for image analysis: label detection, OCR, explicit content detection, and landmark/logo recognition.

Question 29

What is MLOps?

a) Managing ML models using DevOps principles
b) A new ML algorithm
c) An ML certification
d) A cloud service

Show Answer

Answer: a) Managing ML models using DevOps principles

Explanation: MLOps applies DevOps practices (CI/CD, automation, monitoring, versioning) to machine learning, enabling reliable and efficient ML lifecycle management.

Question 30

Which AWS SageMaker feature enables automatic model tuning?

a) SageMaker Automatic Model Tuning (Hyperparameter Optimization)
b) SageMaker AutoML
c) SageMaker Experiments
d) SageMaker Pipelines

Show Answer

Answer: a) SageMaker Automatic Model Tuning (Hyperparameter Optimization)

Explanation: SageMaker Automatic Model Tuning runs multiple training jobs with different hyperparameter combinations to find the best performing model.

Question 31

Which Azure service provides automated ML (AutoML)?

a) Azure Machine Learning (AutoML)
b) Azure Cognitive Services
c) Azure ML Studio
d) Azure Data Science VM

Show Answer

Answer: a) Azure Machine Learning (AutoML)

Explanation: Azure AutoML automatically explores various algorithms and hyperparameters to find the best model for your data, handling feature engineering and model selection.

Question 32

Which GCP service provides AutoML capabilities for custom model training?

a) Vertex AI AutoML
b) Cloud AutoML (legacy)
c) AI Platform
d) Cloud ML

Show Answer

Answer: a) Vertex AI AutoML

Explanation: Vertex AI AutoML (successor to Cloud AutoML) enables training custom ML models for tabular, image, text, and video data with minimal ML expertise.

Question 33

What is the purpose of feature stores in ML?

a) Storing feature definitions only
b) Centralized repository for managing, sharing, and reusing ML features
c) Storing model artifacts
d) Data visualization

Show Answer

Answer: b) Centralized repository for managing, sharing, and reusing ML features

Explanation: Feature stores (like SageMaker Feature Store, Feast) provide a centralized platform for storing, sharing, and serving ML features, ensuring consistency between training and inference.

Question 34

Which AWS service provides a managed ML workflow orchestration?

a) SageMaker Pipelines
b) AWS Step Functions
c) AWS Glue
d) Amazon MWAA

Show Answer

Answer: a) SageMaker Pipelines

Explanation: SageMaker Pipelines creates and manages ML workflows (data processing, training, evaluation, deployment) as DAGs (Directed Acyclic Graphs) with automatic orchestration.

Question 35

What is model drift in MLOps?

a) Model accuracy improving over time
b) Model performance degrading over time due to data changes
c) Model becoming too complex
d) Model being retrained automatically

Show Answer

Answer: b) Model performance degrading over time due to data changes

Explanation: Model drift occurs when the statistical properties of the target variable or input data change over time, causing model predictions to become less accurate. Monitoring for drift enables timely retraining.


Section 3: Deep Learning & Advanced Topics (Questions 36–50)

Question 36

What is a neural network?

a) A biological network of neurons
b) A computing system inspired by biological neural networks, consisting of layers of interconnected nodes
c) A type of database
d) A network protocol

Show Answer

Answer: b) A computing system inspired by biological neural networks, consisting of layers of interconnected nodes

Explanation: Artificial neural networks consist of input, hidden, and output layers of interconnected nodes (neurons) that learn patterns through weighted connections and activation functions.

Question 37

What is the role of an activation function in a neural network?

a) Initializing weights
b) Introducing non-linearity into the network
c) Reducing the number of parameters
d) Normalizing inputs

Show Answer

Answer: b) Introducing non-linearity into the network

Explanation: Activation functions (ReLU, Sigmoid, Tanh) introduce non-linearity, allowing neural networks to learn complex patterns. Without them, the network would be a linear model regardless of depth.

Question 38

Which neural network architecture is best for image classification?

a) Recurrent Neural Network (RNN)
b) Convolutional Neural Network (CNN)
c) Transformer
d) Autoencoder

Show Answer

Answer: b) Convolutional Neural Network (CNN)

Explanation: CNNs are designed for processing grid-like data (images). They use convolutional layers to detect spatial patterns (edges, textures, objects) and pooling layers for downsampling.

Question 39

Which neural network architecture is best for sequence data like text or time series?

a) CNN
b) RNN (or LSTM/GRU)
c) GAN
d) ResNet

Show Answer

Answer: b) RNN (or LSTM/GRU)

Explanation: RNNs (Recurrent Neural Networks) process sequential data by maintaining a hidden state that captures information from previous steps. LSTMs and GRUs address the vanishing gradient problem.

Question 40

What is the Transformer architecture primarily used for?

a) Image generation
b) Natural language processing tasks
c) Clustering
d) Dimensionality reduction

Show Answer

Answer: b) Natural language processing tasks

Explanation: Transformers (BERT, GPT) use self-attention mechanisms to process sequential data in parallel, outperforming RNNs for NLP tasks like translation, summarization, and text generation.

Question 41

What is transfer learning?

a) Moving a model to production
b) Using a pre-trained model and fine-tuning it for a new task
c) Transferring data between storage systems
d) Copying model parameters

Show Answer

Answer: b) Using a pre-trained model and fine-tuning it for a new task

Explanation: Transfer learning takes a model pre-trained on a large dataset (e.g., ImageNet) and fine-tunes it on a smaller, domain-specific dataset, requiring less data and training time.

Question 42

What is the purpose of dropout in neural networks?

a) Dropping low-importance features
b) Preventing overfitting by randomly dropping neurons during training
c) Reducing the number of layers
d) Dropping low-accuracy predictions

Show Answer

Answer: b) Preventing overfitting by randomly dropping neurons during training

Explanation: Dropout randomly deactivates a fraction of neurons during each training iteration, preventing neurons from co-adapting too much and acting as a regularization technique.

Question 43

What is a Generative Adversarial Network (GAN)?

a) A network that classifies images
b) Two networks (generator and discriminator) competing against each other
c) A network for sequence prediction
d) A network for clustering

Show Answer

Answer: b) Two networks (generator and discriminator) competing against each other

Explanation: GANs consist of a generator (creates fake data) and a discriminator (tries to distinguish real from fake). They compete, improving each other, and are used for image generation and data augmentation.

Question 44

What is word embedding in NLP?

a) Converting words to fixed-length vectors capturing semantic meaning
b) A dictionary of words
c) A grammar checker
d) A text formatting tool

Show Answer

Answer: a) Converting words to fixed-length vectors capturing semantic meaning

Explanation: Word embeddings (Word2Vec, GloVe) represent words as dense vectors in a continuous vector space where semantically similar words are close to each other.

Question 45

What is the gradient descent algorithm used for?

a) Data preprocessing
b) Minimizing the loss function by updating model weights
c) Feature selection
d) Model evaluation

Show Answer

Answer: b) Minimizing the loss function by updating model weights

Explanation: Gradient descent iteratively adjusts model parameters in the direction that minimizes the loss function, using the gradient (derivative) to determine update direction and magnitude.

Question 46

What is the learning rate in gradient descent?

a) Number of training iterations
b) Step size for weight updates
c) Number of training examples
d) Model accuracy

Show Answer

Answer: b) Step size for weight updates

Explanation: The learning rate controls how much weights are adjusted during each gradient descent step. Too high: overshoots minimum. Too low: slow convergence.

Question 47

What is batch normalization?

a) Normalizing features before training
b) Normalizing layer inputs across each mini-batch to stabilize training
c) Organizing data into batches
d) Normalizing model outputs

Show Answer

Answer: b) Normalizing layer inputs across each mini-batch to stabilize training

Explanation: Batch normalization normalizes the inputs of each layer across the mini-batch, reducing internal covariate shift and enabling higher learning rates and faster convergence.

Question 48

What is a recommender system?

a) A system that recommends ML algorithms
b) A system that predicts user preferences for items
c) A system that recommends cloud services
d) A system that recommends datasets

Show Answer

Answer: b) A system that predicts user preferences for items

Explanation: Recommender systems predict user preferences for items (products, movies, content). Collaborative filtering, content-based filtering, and hybrid approaches are common techniques.

Question 49

What is A/B testing in ML?

a) Comparing two algorithms
b) Comparing two model versions in production with real users
c) Testing two features
d) Comparing two datasets

Show Answer

Answer: b) Comparing two model versions in production with real users

Explanation: A/B testing (split testing) in ML involves deploying two model versions (control and treatment) to different user segments and comparing performance metrics.

Question 50

What is CI/CD in MLOps?

a) Continuous Integration / Continuous Deployment for ML pipelines
b) Constant Improvement / Continuous Development
c) Code Integration / Code Deployment
d) Continuous Inference / Continuous Delivery

Show Answer

Answer: a) Continuous Integration / Continuous Deployment for ML pipelines

Explanation: CI/CD in MLOps automates testing and deployment of ML models: CI validates code and model quality, CD automates deployment to staging/production environments.


How Did You Score?

  • 0–25 correct: Review AI/ML fundamentals and cloud ML services.
  • 26–40 correct: Good foundation. Practice with hands-on labs.
  • 41–50 correct: Ready for certification exams!

Access all AI/ML practice questions →


Related Articles

Bereit, dein Wissen zu testen?

Probiere unsere Übungsprüfungen mit Hunderten von realistischen Fragen aus.

Üben starten →

This site uses essential cookies for Stripe payments. No tracking cookies.