100+ Multiple Choice Questions on Machine Learning, Deep Learning & AI
Q. What is Machine Learning primarily about?
A. Writing explicit rules
B. Using loops in code
C. Training systems on data to learn patterns
D. Creating databases manually
Answer: Training systems on data to learn patterns
Q. Which of the following best describes Machine Learning?
A. Programming by hard-coding rules
B. Using data and algorithms to learn tasks
C. Designing operating systems
D. Building static web pages
Answer: Using data and algorithms to learn tasks
Q. Which of the following is NOT a Machine Learning algorithm?
A. Support Vector Machine (SVM)
B. Decision Tree
C. K-Means
D. Hypertext Transfer Protocol (HTTP)
Answer: Hypertext Transfer Protocol (HTTP)
Q. What is the role of an algorithm in ML?
A. To store data
B. To manually label inputs
C. To extract patterns from raw data
D. To display results
Answer: To extract patterns from raw data
Q. How is a task learned in ML?
A. By writing if-else rules
B. By coding solutions manually
C. By training models using historical data
D. By replicating database schemas
Answer: By training models using historical data
Q. Which example best explains the concept of “representation” in data?
A. CSV to XLS conversion
B. RGB to HSV image encoding
C. HTTP to HTTPS protocol
D. JPG to PNG compression
Answer: RGB to HSV image encoding
Q. What is the purpose of data representation in ML?
A. To increase file size
B. To make data transmission easier
C. To encode data in a more meaningful form for processing
D. To reduce internet bandwidth
Answer: To encode data in a more meaningful form for processing
Q. Which statement best describes the ML workflow?
A. A set of database scripts
B. Static HTML pages
C. A step-by-step process from data input to prediction
D. A log file analyzer
Answer: A step-by-step process from data input to prediction
Q. In the ML workflow, models are used to:
A. Design websites
B. Encrypt data
C. Make predictions on unseen data
D. Store documents
Answer: Make predictions on unseen data
Q. ML models learn from:
A. Unavailable memory blocks
B. Game engines
C. Historical training data
D. Application caches
Answer: Historical training data
Q. Deep Learning is a subfield of:
A. Robotics
B. Cybersecurity
C. Machine Learning
D. Databases
Answer: Machine Learning
Q. What does “deep” refer to in Deep Learning?
A. Complex memory
B. Deep folders
C. Multiple layers of representation
D. Logical programming
Answer: Multiple layers of representation
Q. Deep Learning uses which type of models?
A. Decision Trees
B. Linear Regression
C. Neural Networks
D. Random Forests
Answer: Neural Networks
Q. A “deep” model usually has:
A. Less than 2 layers
B. Only one hidden layer
C. Multiple successive layers
D. Static input-output mapping
Answer: Multiple successive layers
Q. The depth of a model in DL refers to:
A. The size of the dataset
B. The number of neurons
C. The number of transformation layers
D. The type of activation function
Answer: The number of transformation layers
Q. Which analogy is used to describe how DL “unfolds” data manifolds?
A. Folding a paper plane
B. Crumpling a page
C. Uncrumpling a paper ball
D. Burning a document
Answer: Uncrumpling a paper ball
Q. Each layer in a deep network is responsible for:
A. Deleting data
B. Compressing storage
C. Learning more abstract features
D. Displaying visual charts
Answer: Learning more abstract features
Q. Deep Learning is most suitable for:
A. Small datasets
B. Rule-based systems
C. Large, high-dimensional datasets
D. Text editors
Answer: Large, high-dimensional datasets
Q. Which statement is TRUE about shallow networks?
A. They contain hundreds of layers
B. They require GPU
C. They contain one or two layers
D. They are not used in ML
Answer: They contain one or two layers
Q. The features in DL models are:
A. Manually designed
B. Pre-labeled
C. Automatically learned
D. Encoded in XML
Q. Machine Learning performs well with:
A. Big data only
B. Small to medium datasets
C. No data
D. Textbooks
Answer: Small to medium datasets
Q. Deep Learning requires:
A. Typewriters
B. Spreadsheet tools
C. Powerful machines and GPUs
D. 2G Internet speed
Answer: Powerful machines and GPUs
Q. In ML, feature engineering is:
A. Automatic
B. Unnecessary
C. Manual and crucial
D. Always outsourced
Answer: Manual and crucial
Q. In DL, feature extraction is:
A. Done by hand
B. Skipped
C. Automated by neural networks
D. Limited to images
Answer: Automated by neural networks
Q. Which ML algorithms are easy to interpret?
A. SVM and Neural Nets
B. Logistic Regression and Decision Trees
C. CNN and RNN
D. Deep Belief Networks
Answer: Logistic Regression and Decision Trees
Q. Why has DL become popular now?
A. Increased internet charges
B. Declining computing costs and hardware advances
C. Less demand in the job market
D. Less data available
Answer: Declining computing costs and hardware advances
Q. Which type of learning uses labeled data?
A. Supervised Learning
B. Unsupervised Learning
C. Reinforcement Learning
D. Semi-Supervised Learning
Answer: Supervised Learning
Q. Unsupervised learning is applied when:
A. Labels are available
B. Feedback is real-time
C. No labels are provided
D. Data is encrypted
Answer: No labels are provided
Q. Which learning technique is reward-based?
A. Unsupervised Learning
B. Reinforcement Learning
C. Supervised Learning
D. Symbolic Learning
Answer: Reinforcement Learning
Q. In reinforcement learning, an agent learns by:
A. Passive watching
B. Being punished only
C. Rewards and punishments
D. Random guessing
Answer: Rewards and punishments
Q. Which learning paradigm is best for self-driving cars?
A. Reinforcement Learning
B. Unsupervised Learning
C. Static Rules
D. Symbolic Reasoning
Answer: Reinforcement Learning
Q. What does CRISP-DM stand for in ML project lifecycle?
A. Cross-Industry Standard Process for Data Mining
B. Common Research in Supervised Programming
C. Credit Risk Standard Platform
D. Core Regression Statistics Programming
Answer: Cross-Industry Standard Process for Data Mining
Q. AI (not ML) systems are typically:
A. Rule-based systems
B. Data-driven systems
C. Feedback-oriented systems
D. Exploratory systems
Answer: Rule-based systems
Q. What is one key reason why DL outperforms ML on big data?
A. Requires fewer layers
B. Reduces cost
C. Automates feature engineering
D. Needs no GPU
Answer: Automates feature engineering
Q. In the analogy of a child cleaning their room for a toy, what represents the reward?
A. Cleaning action
B. The toy
C. The room
D. Parent instructions
Answer: The toy
Q. What chemical system in the brain relates to reinforcement learning?
A. Melatonin
B. Serotonin
C. Dopamine
D. Cortisol
Answer: Dopamine
Q. Which ML algorithm is most suitable for classification problems?
A. Support Vector Machine (SVM)
B. K-Means
C. Principal Component Analysis
D. Linear Discriminant Analysis
Answer: Support Vector Machine (SVM)
Q. Which algorithm is best used for clustering problems?
A. Logistic Regression
B. Decision Tree
C. K-Means
D. Naive Bayes
Answer: K-Means
Q. Which ML algorithm splits data into branches based on decision rules?
A. Decision Tree
B. SVM
C. KNN
D. LSTM
Answer: Decision Tree
Q. Which algorithm calculates boundaries between data classes?
A. KNN
B. Decision Tree
C. Support Vector Machine
D. Naive Bayes
Answer: Support Vector Machine
Q. In K-Means, ‘K’ refers to:
A. Number of inputs
B. Number of classes
C. Number of clusters
D. Number of outputs
Answer: Number of clusters
Q. Which algorithm is NOT a classification algorithm?
A. Logistic Regression
B. Naive Bayes
C. K-Means
D. SVM
Answer: K-Means
Q. Which of the following can be used for both classification and regression?
A. Decision Tree
B. K-Means
C. Apriori
D. KNN
Answer: Decision Tree
Q. Which algorithm is commonly used in spam email detection?
A. Naive Bayes
B. K-Means
C. KNN
D. PCA
Answer: Naive Bayes
Q. Which algorithm assumes feature independence in probabilistic classification?
A. Naive Bayes
B. Decision Tree
C. SVM
D. K-Means
Answer: Naive Bayes
Q. A basic unit of a neural network is called a:
A. Block
B. Neuron
C. Pipe
D. Gate
Answer: Neuron
Q. Which function is typically used to introduce non-linearity in DL?
A. Addition
B. Activation Function
C. Subtraction
D. Normalization
Answer: Activation Function
Q. Which activation function is most commonly used in hidden layers?
A. Sigmoid
B. Tanh
C. ReLU (Rectified Linear Unit)
D. Linear
Answer: ReLU (Rectified Linear Unit)
Q. What does CNN stand for in Deep Learning?
A. Central Neural Network
B. Convolutional Neural Network
C. Core Network Node
D. Critical Node Network
Answer: Convolutional Neural Network
Q. CNNs are mainly used for:
A. Image data
B. Text data
C. Tabular data
D. Audio signals
Answer: Image data
Q. What does RNN stand for in Deep Learning?
A. Reversible Neural Network
B. Regularized Neural Network
C. Recurrent Neural Network
D. Random Node Network
Answer: Recurrent Neural Network
Q. RNNs are most suitable for:
A. Predicting tabular data
B. Static classification tasks
C. Sequential data like time-series and text
D. Clustering data
Answer: Sequential data like time-series and text
Q. Which neural network is most used in natural language processing?
A. CNN
B. RNN
C. KNN
D. Decision Trees
Answer: RNN
Q. What are weights in a neural network?
A. Number of neurons
B. Values that determine connection strength
C. Data inputs
D. Labels
Answer: Values that determine connection strength
Q. Which metric is used to evaluate classification models?
A. Mean Squared Error
B. Accuracy
C. R-Squared
D. Euclidean Distance
Answer: Accuracy
Q. Which metric is preferred for imbalanced datasets?
A. Accuracy
B. Confusion Matrix
C. Precision, Recall, F1-Score
D. Mean Absolute Error
Answer: Precision, Recall, F1-Score
Q. Which metric is best for regression problems?
A. Accuracy
B. Precision
C. Mean Squared Error (MSE)
D. Confusion Matrix
Answer: Mean Squared Error (MSE)
Q. AUC-ROC curve is used for evaluating:
A. Regression Models
B. Clustering Algorithms
C. Binary Classification Models
D. Dimensionality Reduction
Answer: Binary Classification Models
Q. What does a confusion matrix represent?
A. Memory usage
B. Classification results in a matrix form
C. Model weights
D. Feature maps
Answer: Classification results in a matrix form
Q. PCA stands for:
A. Principal Component Analysis
B. Pattern Clustering Algorithm
C. Partial Calculation Algorithm
D. Primary Classification Algorithm
Answer: Principal Component Analysis
Q. What does PCA do in ML?
A. Increases dimensionality
B. Reduces features while retaining variance
C. Removes missing values
D. Normalizes data
Answer: Reduces features while retaining variance
Q. Data preprocessing includes:
A. Adding labels
B. Cleaning, scaling, and transforming data
C. Testing models
D. Writing UI code
Answer: Cleaning, scaling, and transforming data
Q. Which scaling technique is commonly used before training models?
A. Zero Padding
B. Normalization or Standardization
C. Flattening
D. Shuffling
Answer: Normalization or Standardization
Q. Which technique handles missing values in data?
A. Data Augmentation
B. Imputation
C. Regularization
D. Noise Injection
Answer: Imputation
Q. Which process transforms text into numerical values in NLP?
A. Parsing
B. Tokenization
C. Vectorization
D. Feature Crossing
Answer: Vectorization
Q. Which framework is commonly used for deep learning in Python?
A. Flask
B. NumPy
C. TensorFlow
D. Selenium
Answer: TensorFlow
Q. Which library is widely used with TensorFlow for high-level APIs?
A. OpenCV
B. Keras
C. Scrapy
D. Django
Answer: Keras
Q. Which company developed TensorFlow?
A. Microsoft
B. Amazon
C. Google
D. IBM
Answer: Google
Q. Which technique helps reduce overfitting in neural networks?
A. Oversampling
B. Dropout
C. SMOTE
D. Feature scaling
Answer: Dropout
Q. What is the key advantage of ensemble methods like Random Forest?
A. Higher interpretability
B. Lower memory usage
C. Better accuracy via multiple models
D. Simple computation
Answer: Better accuracy via multiple models
Q. Which ML model combines weak learners to form a strong learner?
A. Logistic Regression
B. AdaBoost
C. KNN
D. CNN
Answer: AdaBoost
Q. Which learning approach involves learning from human-labeled examples?
A. Unsupervised Learning
B. Reinforcement Learning
C. Supervised Learning
D. Semi-Supervised Learning
Answer: Supervised Learning
Q. Which optimizer is most widely used in deep learning today?
A. Stochastic Gradient Descent
B. Adam
C. Newton’s Method
D. L-BFGS
Answer: Adam
Q. Which loss function is often used for binary classification?
A. Mean Squared Error
B. Categorical Cross-Entropy
C. Binary Cross-Entropy
D. Hinge Loss
Answer: Binary Cross-Entropy
Q. In reinforcement learning, what is an "agent"?
A. The environment in which learning takes place
B. The evaluator of the reward
C. The decision-making entity
D. The dataset used for learning
Answer: The decision-making entity
Q. What defines the environment in reinforcement learning?
A. The algorithm used
B. The agent's actions
C. Everything the agent interacts with
D. Only the reward system
Answer: Everything the agent interacts with
Q. What are rewards in reinforcement learning?
A. Values that represent success or failure of the agent’s action
B. Parameters used for optimization
C. Penalties applied to actions
D. Layers in a deep learning model
Answer: Values that represent success or failure of the agent’s action
Q. The mapping of states and actions to expected rewards is known as:
A. Loss function
B. Transition matrix
C. Value function
D. Policy gradient
Answer: Value function
Q. Which of the following is an example of reinforcement learning in the real world?
A. Predicting house prices
B. Classifying emails
C. A robot learning to walk
D. Finding customer segments
Answer: A robot learning to walk
Q. Which chemical in the brain is linked to reinforcement behavior?
A. Serotonin
B. Dopamine
C. Adrenaline
D. Oxytocin
Answer: Dopamine
Q. In reinforcement learning, what is a "policy"?
A. Rules followed by the environment
B. A model that maps actions to states
C. A mapping from states to actions
D. A set of rewards
Answer: A mapping from states to actions
Q. What does CRISP-DM stand for?
A. Common Reasoning in Supervised Projects - Data Modeling
B. Cross-Industry Standard Process for Data Mining
C. Cloud-Ready Integrated System for Predictive Data Modeling
D. Central Repository for Intelligent Statistical Predictions
Answer: Cross-Industry Standard Process for Data Mining
Q. Which is the first step in CRISP-DM?
A. Data modeling
B. Deployment
C. Business understanding
D. Data cleaning
Answer: Business understanding
Q. What is the final stage in the CRISP-DM framework?
A. Data validation
B. Deployment
C. Business understanding
D. Evaluation
Answer: Deployment
Q. What does the "data preparation" phase include?
A. Building machine learning models
B. Creating test cases
C. Cleaning and transforming data
D. Deploying models into production
Answer: Cleaning and transforming data
Q. Which ML models are considered interpretable?
A. Decision Trees and Logistic Regression
B. Neural Networks and SVM
C. Gradient Boosting Machines
D. Deep Belief Networks
Answer: Decision Trees and Logistic Regression
Q. Why is model interpretability important?
A. To reduce runtime
B. To improve data compression
C. To understand how predictions are made
D. To increase dataset size
Answer: To understand how predictions are made
Q. Which model is often considered a "black-box" model?
A. Decision Tree
B. K-Nearest Neighbors
C. Deep Neural Network
D. Logistic Regression
Answer: Deep Neural Network
Q. Which of the following is a concern in ML ethics?
A. Model compression
B. Bias and fairness in prediction
C. Hyperparameter tuning
D. Model accuracy
Answer: Bias and fairness in prediction
Q. What is data anonymization?
A. Removing training data
B. Encrypting weights
C. Obscuring personal identifiers in data
D. Repeating model evaluation
Answer: Obscuring personal identifiers in data
Q. Which hardware is essential for training deep learning models efficiently?
A. Hard Drive
B. CPU
C. SSD
D. GPU
Answer: GPU
Q. Which of the following is a deep learning platform by Facebook?
A. TensorFlow
B. PyTorch
C. Keras
D. Scikit-learn
Answer: PyTorch
Q. Which ML library is best for classical algorithms like SVM or Decision Tree?
A. Keras
B. TensorFlow
C. Scikit-learn
D. OpenCV
Answer: Scikit-learn
Q. Which file format is commonly used for datasets?
A. .exe
B. .mp4
C. .csv
D. .html
Answer: .csv
Q. Which language is most commonly used in ML projects?
A. C++
B. Java
C. Python
D. Ruby
Answer: Python
Q. What makes deep learning more popular today than in the past?
A. Smaller datasets
B. Better manual features
C. More computing power and data
D. Fewer use cases
Answer: More computing power and data
Q. Which innovation supports DL by reducing cost of training?
A. CD-ROM
B. Low-end CPUs
C. GPUs and TPUs
D. Ethernet cables
Answer: GPUs and TPUs
Q. Which field benefits the most from computer vision and DL?
A. Agriculture
B. Image recognition
C. Blogging
D. Networking
Answer: Image recognition
Q. Which of the following is not a subfield of machine learning?
A. Supervised Learning
B. Unsupervised Learning
C. GraphQL
D. Reinforcement Learning
Answer: GraphQL
Q. Which method is used when labels are not available?
A. Supervised Learning
B. Semi-supervised Learning
C. Reinforcement Learning
D. Unsupervised Learning
Answer: Unsupervised Learning
Q. What makes deep learning models computationally expensive?
A. They use fewer neurons
B. They involve shallow representations
C. They perform heavy matrix computations
D. They use simple logic
Answer: They perform heavy matrix computations
Q. Which learning type requires exploration and reward maximization?
A. Reinforcement Learning
B. Supervised Learning
C. Semi-supervised Learning
D. Unsupervised Learning
Answer: Reinforcement Learning