14 Sep 100 Machine Learning MCQ (Multiple Choice Questions) with Answers

Posted at 18:10h in mcq by

1) What is Machine Learning?
  1. A branch of AI that allows systems to learn from data without explicit programming
  2. A method of manually writing rules for every scenario
  3. A hardware architecture for supercomputers
  4. A database management system
Show Answer
Answer: a
Explanation
Machine Learning is a subset of AI where models learn patterns from data rather than being explicitly programmed with rules for every scenario.

2) Which of the following is a type of Supervised Learning?

  1. K-Means Clustering
  2. Linear Regression
  3. Apriori Algorithm
  4. Principal Component Analysis (PCA)
Show Answer
Answer: b
Explanation
Linear Regression is a supervised learning algorithm because it learns from labeled data (input-output pairs). The others are unsupervised techniques.

3) What type of target variable does a Classification problem have?

  1. Continuous
  2. Categorical
  3. Complex numbers
  4. Infinite numerical values
Show Answer
Answer: b
Explanation
Classification predicts discrete categories or classes (e.g., spam/not spam), so the target variable is categorical.

4) What type of target variable does a Regression problem have?

  1. Binary categories
  2. Ordinal text
  3. Continuous numerical values
  4. Unlabeled clusters
Show Answer
Answer: c
Explanation
Regression predicts a continuous numeric quantity (e.g., house price, temperature), so the target is continuous.

5) Which algorithm is commonly used for unsupervised clustering?

  1. Decision Tree
  2. Support Vector Machine
  3. K-Means
  4. Logistic Regression
Show Answer
Answer: c
Explanation
K-Means is an unsupervised clustering algorithm that groups data points into K clusters based on similarity.

6) What is the primary cause of overfitting in a machine learning model?

  1. Model is too simple
  2. Model learned noise and details of the training data instead of general patterns
  3. Dataset is too large
  4. Learning rate is too high
Show Answer
Answer: b
Explanation
Overfitting occurs when a model memorizes noise and specific details of the training set, failing to generalize to new data.

7) How can you identify underfitting in a model?

  1. High performance on training set, low performance on test set
  2. Low performance on both training set and test set
  3. High performance on both training set and test set
  4. Perfect accuracy on validation set
Show Answer
Answer: b
Explanation
Underfitting means the model is too simple to capture the underlying patterns, resulting in poor performance on both training and test data.

8) Which technique helps prevent overfitting by adding a penalty term to the loss function?

  1. Normalization
  2. Regularization
  3. Imputation
  4. One-Hot Encoding
Show Answer
Answer: b
Explanation
Regularization (L1/L2) adds a penalty to the loss function to discourage complex models, reducing overfitting.

9) What is L1 Regularization also known as?

  1. Ridge Regularization
  2. Lasso Regularization
  3. ElasticNet
  4. Dropout
Show Answer
Answer: b
Explanation
L1 regularization is called Lasso; it adds the sum of absolute weights and can drive some coefficients to zero.

10) What is L2 Regularization also known as?

  1. Lasso Regularization
  2. Ridge Regularization
  3. Gradient Descent
  4. Min-Max Scaling
Show Answer
Answer: b
Explanation
L2 regularization is called Ridge; it adds the sum of squared weights and shrinks coefficients smoothly.

11) Which metric is used to evaluate a classification model when classes are highly imbalanced?

  1. Mean Squared Error
  2. F1-Score
  3. R-Squared
  4. Mean Absolute Error
Show Answer
Answer: b
Explanation
The F1-Score balances precision and recall, making it suitable for imbalanced datasets where accuracy can be misleading.

12) In a Confusion Matrix, what does a “False Positive” mean?

  1. Predicting a positive class correctly
  2. Predicting a negative class correctly
  3. Predicting a positive class when the actual class was negative
  4. Predicting a negative class when the actual class was positive
Show Answer
Answer: c
Explanation
A False Positive is when the model predicts the positive class but the true label is negative (Type I error).

13) What is Precision?

  1. True Positives / (True Positives + False Positives)
  2. True Positives / (True Positives + False Negatives)
  3. True Negatives / (True Negatives + False Positives)
  4. Total Correct Predictions / Total Predictions
Show Answer
Answer: a
Explanation
Precision measures how many of the predicted positive cases are actually positive.

14) What is Recall (Sensitivity)?

  1. True Positives / (True Positives + False Positives)
  2. True Positives / (True Positives + False Negatives)
  3. False Positives / (True Negatives + False Positives)
  4. True Negatives / Total Predictions
Show Answer
Answer: b
Explanation
Recall measures how many of the actual positive cases were correctly identified by the model.

15) What is the F1-Score?

  1. Arithmetic Mean of Precision and Recall
  2. Geometric Mean of Precision and Recall
  3. Harmonic Mean of Precision and Recall
  4. Median of Precision and Recall
Show Answer
Answer: c
Explanation
F1-Score is the harmonic mean of Precision and Recall, giving a balanced measure that penalizes extreme values.

16) Which algorithm uses a hyper-plane to separate data points into classes?

  1. Linear Regression
  2. Support Vector Machine (SVM)
  3. Naïve Bayes
  4. K-Means
Show Answer
Answer: b
Explanation
SVM finds the optimal hyperplane that maximally separates classes in the feature space.

17) What is the main assumption of the Naïve Bayes classifier?

  1. Features are highly correlated
  2. Features are independent of each other given the class label
  3. Data follows a uniform distribution
  4. Target variable must be continuous
Show Answer
Answer: b
Explanation
Naïve Bayes assumes conditional independence of features given the class, which simplifies computation.

18) Which metric measures impurity in a Decision Tree?

  1. Mean Squared Error
  2. Gini Impurity
  3. Euclidean Distance
  4. Cosine Similarity
Show Answer
Answer: b
Explanation
Gini Impurity measures how often a randomly chosen element would be incorrectly classified; lower values indicate purer nodes.

19) Ensemble method that builds multiple decision trees in parallel using bootstrap samples is called:

  1. Gradient Boosting
  2. AdaBoost
  3. Random Forest
  4. Stacking
Show Answer
Answer: c
Explanation
Random Forest builds many decision trees in parallel on bootstrap samples and aggregates their predictions.

20) What is the main difference between Bagging and Boosting?

  1. Bagging trains models sequentially; Boosting trains them in parallel
  2. Bagging trains models in parallel; Boosting trains them sequentially
  3. Bagging is only for regression; Boosting is only for classification
  4. Bagging reduces bias; Boosting only reduces variance
Show Answer
Answer: b
Explanation
Bagging trains base models independently in parallel, while Boosting trains them sequentially, each focusing on previous errors.

21) Gradient Boosting belongs to which family of algorithms?

  1. Unsupervised Learning
  2. Boosting Ensembles
  3. Reinforcement Learning
  4. Dimensionality Reduction
Show Answer
Answer: b
Gradient Boosting is a boosting ensemble method that builds models sequentially to correct errors of prior models.

22) What is Principal Component Analysis (PCA) used for?

  1. Classification
  2. Dimensionality Reduction
  3. Handling Missing Values
  4. Time Series Forecasting
Show Answer
Answer: b
Explanation
PCA is an unsupervised technique that reduces dimensionality by projecting data onto principal components.

23) In K-Means clustering, what does ‘K’ represent?

  1. Number of features
  2. Number of iterations
  3. Number of clusters
  4. Number of nearest neighbors
Show Answer
Answer: c
Explanation
In K-Means, K is the number of clusters the algorithm should find.

24) In KNN (K-Nearest Neighbors), what does ‘K’ represent?

  1. Number of output classes
  2. Number of training samples
  3. Number of nearest neighbors to consider for classification/regression
  4. Number of decision trees
Show Answer
Answer: c
Explanation
In KNN, K is the number of nearest neighbors used to make a prediction.

25) Which distance metric is most commonly used in default KNN?

  1. Manhattan Distance
  2. Cosine Distance
  3. Euclidean Distance
  4. Hamming Distance
Show Answer
Answer: c
Explanation
Euclidean distance is the default distance metric in most KNN implementations.

26) What is the Curse of Dimensionality?

  1. Difficulty in storing high-dimensional arrays
  2. Decrease in model performance and computational efficiency as feature dimensions increase dramatically
  3. Overfitting caused by having too few features
  4. Gradient vanishing during neural network training
Show Answer
Answer: b
Explanation
As dimensionality increases, data becomes sparse, making it harder for models to learn and increasing computational cost.

27) What is the purpose of cross-validation (e.g., K-Fold)?

  1. To increase training time
  2. To evaluate model performance reliably and avoid overfitting
  3. To compress the dataset
  4. To convert continuous features into categorical ones
Show Answer
Answer: b
Explanation
Cross-validation provides a more robust estimate of model performance by testing on multiple folds of the data.

28) In K-Fold Cross Validation with K=5, how many times is the model trained?

  1. 1
  2. 4
  3. 5
  4. 25
Show Answer
Answer: c
Explanation
With K=5, the model is trained 5 times, each time using a different fold as validation.

29) What is One-Hot Encoding used for?

  1. Scaling continuous numerical features
  2. Converting categorical variables into binary vectors
  3. Removing outliers from data
  4. Imputing missing values
Show Answer
Answer: b
Explanation
One-Hot Encoding creates binary columns for each category, allowing models to use categorical data.

30) Which activation function scales outputs between 0 and 1, often used in binary classification output layers?

  1. ReLU
  2. Sigmoid
  3. Tanh
  4. Leaky ReLU
Show Answer
Answer: b
Explanation
Sigmoid squashes outputs to (0, 1), making it suitable for binary classification probabilities.

31) What is the output range of the Tanh activation function?

  1. [0, 1]
  2. [-1, 1]
  3. [0, infinity)
  4. (-infinity, infinity)
Show Answer
Answer: b
Explanation
Tanh outputs values between -1 and 1, zero-centered, which often helps training.

32) What issue does the Rectified Linear Unit (ReLU) activation function suffer from?

  1. Vanishing Gradient for positive values
  2. Dying ReLU (neurons becoming permanently inactive for negative values)
  3. High computational overhead
  4. Output limited to [0, 1]
Show Answer
Answer: b
Explanation
ReLU can cause neurons to “die” if they consistently output zero for negative inputs, preventing weight updates.

33) What optimization algorithm updates parameters by calculating gradients on small random subsets of data?

  1. Batch Gradient Descent
  2. Stochastic Gradient Descent (SGD) / Mini-Batch SGD
  3. Grid Search
  4. Genetic Algorithm
Show Answer
Answer: b
Explanation
SGD uses small random subsets (mini-batches) to update parameters, making it faster and more scalable.

34) What is the learning rate in Gradient Descent?

  1. The time it takes to train a model
  2. Hyperparameter that controls the step size at each iteration toward a minimum
  3. The percentage of correct predictions
  4. Number of epochs
Show Answer
Answer: b
Explanation
The learning rate determines how much the model’s weights are adjusted during each update step.

35) What happens if the learning rate is too high in Gradient Descent?

  1. The model converges very slowly
  2. The model may overshoot the minimum and diverge
  3. The model stops updating immediately
  4. The loss function drops to zero instantly
Show Answer
Answer: b
Explanation
A high learning rate can cause the optimizer to overshoot the minimum and fail to converge.

36) What happens if the learning rate is too low in Gradient Descent?

  1. Training will be extremely slow and might get stuck in local minima
  2. Model diverges immediately
  3. Accuracy drops to 0%
  4. The weights become infinite
Show Answer
Answer: a
Explanation
A very small learning rate makes training slow and may get trapped in local minima.

37) Which evaluation metric is appropriate for Linear Regression?

  1. Log Loss
  2. Mean Absolute Error (MAE)
  3. Accuracy
  4. ROC-AUC
Show Answer
Answer: b
Explanation
MAE measures the average absolute difference between predicted and actual values, suitable for regression.

38) What does R-squared (R²) represent in regression analysis?

  1. Proportion of variance in the dependent variable explained by independent variables
  2. Absolute error average
  3. Ratio of true positives to false positives
  4. Probability of binary classification
Show Answer
Answer: a
Explanation
R² indicates how well the regression line fits the data, ranging from 0 to 1.

39) What is Logistic Regression primarily used for?

  1. Continuous numeric estimation
  2. Classification
  3. Image generation
  4. Clustering
Show Answer
Answer: b
Explanation
Despite its name, Logistic Regression is used for binary classification by estimating probabilities.

40) What function converts linear model output to probability in Logistic Regression?

  1. Softmax
  2. Sigmoid
  3. Exponential
  4. Polynomial
Show Answer
Answer: b
Explanation
The sigmoid function maps any real value to a probability between 0 and 1.

41) Which algorithm is non-parametric and instance-based?

  1. Linear Regression
  2. K-Nearest Neighbors (KNN)
  3. Naïve Bayes
  4. Logistic Regression
Show Answer
Answer: b
Explanation
KNN is non-parametric and instance-based because it makes predictions based on the nearest training instances.

42) What is a hyperparameter?

  1. A parameter learned by the model during training
  2. A configuration setting set before starting the model training process
  3. The target variable
  4. The error rate on the test set
Show Answer
Answer: b
Explanation
Hyperparameters are set before training (e.g., learning rate, number of trees) and are not learned from data.

43) Which technique automates searching across a pre-defined grid of hyperparameters?

  1. Gradient Descent
  2. Grid Search CV
  3. Principal Component Analysis
  4. Feature Scaling
Show Answer
Answer: b
Explanation
Grid Search CV exhaustively tries all combinations of hyperparameters in a specified grid.

44) What is the purpose of Data Normalization / Standard Scaling?

  1. To remove missing values
  2. To bring all features to a similar numerical scale
  3. To increase the number of features
  4. To label unlabeled data
Show Answer
Answer: b
Explanation
Scaling ensures that features contribute equally to the model, improving convergence and performance.

45) Min-Max Scaling transforms data to which range by default?

  1. [-1, 1]
  2. [0, 1]
  3. [0, 100]
  4. [-infinity, infinity]
Show Answer
Answer: b
Explanation
Min-Max scaling linearly transforms features to the range [0, 1].

46) Standardization (Z-score scaling) transforms data to have:

  1. Mean = 0, Standard Deviation = 1
  2. Mean = 1, Standard Deviation = 0
  3. Min = 0, Max = 1
  4. Mean = 0, Variance = 0
Show Answer
Answer: a
Explanation
Standardization centers data to mean 0 and scales to unit variance.

47) What is Data Imputation?

  1. Removing duplicate rows
  2. Replacing missing values with estimated values (e.g., mean, median)
  3. Converting numbers to text
  4. Splitting data into train/test sets
Show Answer
Answer: b
Explanation
Imputation fills missing data with reasonable estimates to avoid losing records.

48) What is the bias-variance tradeoff?

  1. Balancing model simplicity/underfitting (high bias) against model complexity/overfitting (high variance)
  2. Balancing precision and recall
  3. Balancing training speed and testing speed
  4. Balancing dataset size and feature counts
Show Answer
Answer: a
Explanation
The tradeoff is between a model that is too simple (high bias) and one that is too complex (high variance).

49) High bias usually leads to:

  1. Overfitting
  2. Underfitting
  3. Perfect accuracy
  4. Zero training error
Show Answer
Answer: b
Explanation
High bias means the model is too simple, leading to underfitting.

50) High variance usually leads to:

  1. Underfitting
  2. Overfitting
  3. Generalization
  4. High training error
Show Answer
Answer: b
Explanation
High variance means the model is too sensitive to training data, leading to overfitting.

51) Which loss function is commonly used in binary classification?

  1. Mean Squared Error
  2. Binary Cross-Entropy (Log Loss)
  3. Huber Loss
  4. Mean Absolute Percentage Error
Show Answer
Answer: b
Explanation
Binary Cross-Entropy measures the difference between predicted probabilities and actual binary labels.

52) What is the Softmax activation function used for?

  1. Regression outputs
  2. Multi-class classification probability outputs
  3. Dimensionality reduction
  4. Scaling input features
Show Answer
Answer: b
Explanation
Softmax converts a vector of raw scores into probabilities that sum to 1 for multi-class classification.

53) In Decision Trees, what is “pruning”?

  1. Adding more nodes to the tree
  2. Removing branches to simplify the tree and prevent overfitting
  3. Normalizing input values
  4. Combining multiple trees together
Show Answer
Answer: b
Explanation
Pruning removes weak branches to reduce complexity and improve generalization.

54) What is Bootstrap Aggregating commonly known as?

  1. Boosting
  2. Stacking
  3. Bagging
  4. Blending
Show Answer
Answer: c
Explanation
Bootstrap Aggregating is commonly known as Bagging.

55) Which of the following is a Gradient Boosting implementation?

  1. XGBoost
  2. K-Means
  3. PCA
  4. Apriori
Show Answer
Answer: a
Explanation
XGBoost is a popular and efficient implementation of gradient boosting.

56) What type of machine learning involves an agent interacting with an environment to maximize rewards?

  1. Supervised Learning
  2. Unsupervised Learning
  3. Reinforcement Learning
  4. Semi-supervised Learning
Show Answer
Answer: c
Explanation
Reinforcement Learning involves an agent learning from rewards through interaction with an environment.

57) What is Q-Learning?

  1. A supervised classification algorithm
  2. A model-free reinforcement learning algorithm
  3. A clustering algorithm
  4. A feature extraction technique
Show Answer
Answer: b
Explanation
Q-Learning is a model-free reinforcement learning algorithm that learns the value of actions in states.

58) What is an Outlier?

  1. A missing value in a dataset
  2. A data point that differs significantly from other observations
  3. A category with high frequency
  4. A feature with zero variance
Show Answer
Answer: b
Explanation
An outlier is an observation that lies far from the general distribution of the data.

59) Which visual plot is effective for detecting univariate outliers?

  1. Line plot
  2. Box plot
  3. Confusion Matrix
  4. ROC Curve
Show Answer
Answer: b
Explanation
A box plot visualizes the distribution and highlights points beyond the whiskers as potential outliers.

60) What is the ROC curve?

  1. Plot of True Positive Rate vs. False Positive Rate across thresholds
  2. Plot of Precision vs. Recall
  3. Plot of Loss vs. Epochs
  4. Plot of Variance vs. Bias
Show Answer
Answer: a
Explanation
The ROC curve plots the true positive rate against the false positive rate at various classification thresholds.

61) What does an AUC (Area Under Curve) of 1.0 indicate?

  1. Completely random classifier
  2. Worst possible classifier
  3. Perfect classifier
  4. Overfitted linear regression
Show Answer
Answer: c
Explanation
An AUC of 1.0 means the classifier perfectly distinguishes between classes.

62) An AUC value of 0.5 indicates:

  1. Perfect classification
  2. Performance no better than random guessing
  3. 100% false positive rate
  4. High precision
Show Answer
Answer: b
Explanation
An AUC of 0.5 means the model’s performance is equivalent to random chance.

63) What is the main purpose of dimensionality reduction?

  1. Increase training time
  2. Reduce number of random variables under consideration to simplify data
  3. Generate synthetic samples
  4. Convert unstructured text into numbers
Show Answer
Answer: b
Explanation
Dimensionality reduction simplifies data by reducing the number of features while preserving important information.

64) Which algorithm is an unsupervised density-based clustering method?

  1. K-Means
  2. Hierarchical Clustering
  3. DBSCAN
  4. Support Vector Regression
Show Answer
Answer: c
Explanation
DBSCAN is a density-based clustering algorithm that groups points based on density.

65) What advantage does DBSCAN have over K-Means?

  1. Faster execution time always
  2. Does not require specifying the number of clusters in advance and handles arbitrary shapes
  3. Guaranteed global minimum
  4. Works only on categorical data
Show Answer
Answer: b
Explanation
DBSCAN automatically determines the number of clusters and can find clusters of arbitrary shape.

66) What is a dendrogram used for?

  1. Visualizing Decision Tree splits
  2. Visualizing Hierarchical Clustering structures
  3. Plotting Neural Network weights
  4. Showing ROC curves
Show Answer
Answer: b
Explanation
A dendrogram is a tree-like diagram that shows the arrangement of clusters in hierarchical clustering.

67) In Naïve Bayes, what handles zero probability issues during prediction?

  1. Gradient Descent
  2. Laplace Smoothing
  3. Min-Max Scaling
  4. Early Stopping
Show Answer
Answer: b
Explanation
Laplace Smoothing adds a small constant to avoid zero probabilities in Naïve Bayes.

68) What is Feature Engineering?

  1. Selecting hardware GPUs for training
  2. Process of using domain knowledge to create new features or transform existing ones for better model performance
  3. Automatic hyperparameter tuning
  4. Writing documentation for code
Show Answer
Answer: b
Explanation
Feature engineering uses domain expertise to create informative features that improve model accuracy.

69) What is Multicollinearity?

  1. High correlation between target and feature variables
  2. High correlation among two or more independent (predictor) variables
  3. Lack of data in training set
  4. Presence of multiple classes in output
Show Answer
Answer: b
Explanation
Multicollinearity occurs when predictor variables are highly correlated, which can destabilize regression coefficients.

70) Which metric detects multicollinearity in regression models?

  1. Confusion Matrix
  2. Variance Inflation Factor (VIF)
  3. Silhouette Coefficient
  4. Gini Index
Show Answer
Answer: b
Explanation
VIF measures how much the variance of an estimated regression coefficient is inflated due to collinearity.

71) What does a VIF value greater than 5 or 10 indicate?

  1. Low feature importance
  2. High multicollinearity
  3. High model accuracy
  4. Underfitting
Show Answer
Answer: b
Explanation
VIF above 5 or 10 suggests problematic multicollinearity among predictors.

72) What technique generates synthetic minority class samples to solve class imbalance?

  1. One-Hot Encoding
  2. SMOTE (Synthetic Minority Over-sampling Technique)
  3. Principal Component Analysis
  4. Stratified K-Fold
Show Answer
Answer: b
Explanation
SMOTE creates new synthetic samples for the minority class by interpolating between existing ones.

73) What is Stratified K-Fold cross validation?

  1. K-Fold where fold selection ensures equal class distribution in each fold
  2. Random splitting without replacement
  3. Splitting data based on time intervals
  4. K-Fold without test sets
Show Answer
Answer: a
Explanation
Stratified K-Fold preserves the proportion of classes in each fold, important for imbalanced datasets.

74) What is Early Stopping in iterative models (like neural networks or boosting)?

  1. Stopping training as soon as loss reaches zero
  2. Halting training when validation performance stops improving to prevent overfitting
  3. Stopping training after 1 epoch
  4. Terminating execution when an error occurs
Show Answer
Answer: b
Explanation
Early stopping monitors validation performance and stops training when it begins to degrade.

75) What is the function of Dropout in neural networks?

  1. Speeds up computation by dropping layers
  2. Randomly deactivates neurons during training to prevent overfitting
  3. Deletes rows with missing values
  4. Reduces learning rate automatically
Show Answer
Answer: b
Explanation
Dropout randomly turns off neurons during training, forcing the network to learn more robust features.

76) Which algorithm relies on Bayes’ Theorem?

  1. Random Forest
  2. Support Vector Machine
  3. Naïve Bayes
  4. K-Means
Show Answer
Answer: c
Explanation
Naïve Bayes uses Bayes’ Theorem with an assumption of independence between features.

77) What is the primary metric to evaluate a K-Means clustering quality?

  1. Accuracy
  2. Silhouette Score / Inertia
  3. Mean Squared Error
  4. F1-Score
Show Answer
Answer: b
Explanation
Silhouette Score and Inertia are used to evaluate the quality of clustering without labels.

78) What range does the Silhouette Score span?

  1. [0, 1]
  2. [-1, 1]
  3. [0, 100]
  4. (-infinity, infinity)
Show Answer
Answer: b
Explanation
The Silhouette Score ranges from -1 to 1, where higher values indicate better-defined clusters.

79) A Silhouette Score close to +1 indicates:

  1. Samples are poorly clustered and close to neighboring clusters
  2. Samples are well-clustered and far from neighboring clusters
  3. Wrong number of clusters selected
  4. Data points are outliers
Show Answer
Answer: b
Explanation
A score near +1 means the sample is far from neighboring clusters, indicating good clustering.

80) What is a popular heuristic method to select the optimal number of clusters K in K-Means?

  1. Gradient Descent
  2. Elbow Method
  3. Grid Search
  4. Cross-Validation Score
Show Answer
Answer: b
Explanation
The Elbow Method plots inertia vs. K and looks for the “elbow” point where improvement slows.

81) Which regularizer forces model coefficients strictly to zero, effectively performing feature selection?

  1. L2 Regularization (Ridge)
  2. L1 Regularization (Lasso)
  3. Batch Normalization
  4. Softmax
Show Answer
Answer: b
Explanation
L1 regularization (Lasso) can shrink some coefficients exactly to zero, performing feature selection.

82) What is the cost function for Linear Regression?

  1. Binary Cross-Entropy
  2. Mean Squared Error (MSE)
  3. Categorical Cross-Entropy
  4. Hinge Loss
Show Answer
Answer: b
Explanation
Linear Regression typically minimizes Mean Squared Error between predictions and actual values.

83) What is Hinge Loss commonly used for?

  1. Linear Regression
  2. Support Vector Machines (SVM)
  3. Logistic Regression
  4. Decision Trees
Show Answer
Answer: b
Explanation
Hinge Loss is used in SVMs for maximum-margin classification.

84) In Support Vector Machines, what are Support Vectors?

  1. Feature vectors that lie furthest from the margin
  2. Data points closest to the decision boundary/margin hyper-plane
  3. Vector representations of word counts
  4. Random noise points
Show Answer
Answer: b
Explanation
Support vectors are the data points that lie closest to the decision boundary and influence its position.

85) What is the “Kernel Trick” in SVMs?

  1. A method to skip training
  2. Mapping data into higher-dimensional space to make non-linearly separable data separable
  3. Reducing dimension to 1D
  4. Converting continuous labels into discrete classes
Show Answer
Answer: b
Explanation
The kernel trick implicitly maps data to a higher-dimensional space where a linear separator can be found.

86) Which kernel is widely used in SVM for non-linear data?

  1. Radial Basis Function (RBF)
  2. Linear
  3. Step Function
  4. Absolute Difference
Show Answer
Answer: a
Explanation
The RBF kernel is popular for non-linear SVM because it can handle complex boundaries.

87) What is Data Drift / Covariate Shift?

  1. Change in code syntax across software updates
  2. Change in input data distribution over time compared to training data
  3. Loss of training features
  4. Variance increase due to hardware limits
Show Answer
Answer: b
Explanation
Data drift occurs when the statistical properties of input data change over time, degrading model performance.

88) Transfer Learning refers to:

  1. Moving data from local machine to cloud
  2. Reusing a pre-trained model on a new related task
  3. Transferring labels from train set to test set
  4. Converting code from Python to C++
Show Answer
Answer: b
Explanation
Transfer learning leverages knowledge from a pre-trained model to improve learning on a new but related task.

89) What is semi-supervised learning?

  1. Training using only unlabeled data
  2. Training using a small amount of labeled data and a large amount of unlabeled data
  3. Training without hyperparameter tuning
  4. Alternating between regression and classification
Show Answer
Answer: b
Explanation
Semi-supervised learning combines a small labeled dataset with a large unlabeled dataset.

90) Which gradient descent variant processes the entire dataset in a single step per weight update?

  1. Mini-Batch Gradient Descent
  2. Stochastic Gradient Descent
  3. Batch Gradient Descent
  4. Online Learning
Show Answer
Answer: c
Explanation
Batch Gradient Descent uses the entire training set to compute the gradient for each update.

91) What is an epoch in machine learning training?

  1. Single update of weights using one batch
  2. One complete pass through the entire training dataset
  3. Time duration of 1 hour of training
  4. Number of trees in a forest
Show Answer
Answer: b
Explanation
An epoch is one full pass through all training samples.

92) What is the target variable in Unsupervised Learning?

  1. Binary variable
  2. Continuous variable
  3. There is no target variable
  4. Categorical multi-class variable
Show Answer
Answer: c
Explanation
Unsupervised learning works with unlabeled data, so there is no target variable.

93) Which technique converts ordinal categorical values into numerical ranks (e.g., Low=1, Medium=2, High=3)?

  1. One-Hot Encoding
  2. Label / Ordinal Encoding
  3. Binary Encoding
  4. Mean Encoding
Show Answer
Answer: b
Explanation
Ordinal encoding assigns integers to categories while preserving their order.

94) What is target leakage (data leakage)?

  1. Sharing data with open-source repositories
  2. Information from outside the training dataset (or future test set) being used to create the model
  3. Memory overflow during training
  4. Removing missing values automatically
Show Answer
Answer: b
Explanation
Target leakage occurs when the model uses information not available at prediction time, leading to overly optimistic performance.

95) Which technique is used to find association rules between items in transactional data?

  1. Apriori Algorithm
  2. Support Vector Machine
  3. Linear Discriminant Analysis
  4. Gradient Boosting
Show Answer
Answer: a
Explanation
The Apriori algorithm mines frequent itemsets and generates association rules.

96) Support, Confidence, and Lift are metrics evaluated in:

  1. Time Series Forecasting
  2. Association Rule Mining
  3. Neural Networks
  4. Principal Component Analysis
Show Answer
Answer: b
Explanation
Support, Confidence, and Lift are used to evaluate the strength of association rules.

97) In time series analysis, what is “Stationarity”?

  1. A property where statistical properties like mean and variance remain constant over time
  2. Data points standing still without changes
  3. Seasonal trends that grow exponentially
  4. Absence of missing values
Show Answer
Answer: a
Explanation
Stationarity means the statistical properties of the series do not change over time, a key assumption in many time series models.

98) What model family combines AutoRegression and Moving Average for time series?

  1. Random Forest
  2. ARIMA
  3. K-Means
  4. SVM
Show Answer
Answer: b
Explanation
ARIMA stands for AutoRegressive Integrated Moving Average and is widely used for time series forecasting.

99) What does the “Gradient” in Gradient Descent represent?

  1. Rate of model training speed
  2. Vector of partial derivatives pointing in the direction of steepest increase of the cost function
  3. Accuracy improvement curve
  4. Ratio of bias to variance
Show Answer
Answer: b
Explanation
The gradient points in the direction of steepest ascent; we move opposite to it to minimize the cost.

100) Why is data split into Training and Testing sets?

  1. To make data fit in memory
  2. To evaluate how well the trained model generalizes to unseen data
  3. To double the size of the dataset
  4. Required by Python syntax
Show Answer
Answer: b
Explanation
Splitting data allows you to test the model on unseen data, estimating its real-world performance.
100 Android MCQ (Multiple Choice Questions) with Answers
100 Artificial Intelligence MCQ (Multiple Choice Questions) with Answers
Print page
0 Likes
Studyopedia Editorial Staff
contact@studyopedia.com

We work to create programming tutorials for all.

No Comments

Post A Comment